Spaces:
Running
Running
File size: 2,040 Bytes
9fd1204 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 |
import torch
from diffusers import AutoencoderKLWan, FlowMatchEulerDiscreteScheduler, WanTransformer3DModel
from transformers import AutoTokenizer, T5EncoderModel
from finetrainers.models.wan import WanModelSpecification
class DummyWanModelSpecification(WanModelSpecification):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def load_condition_models(self):
text_encoder = T5EncoderModel.from_pretrained(
"hf-internal-testing/tiny-random-t5", torch_dtype=self.text_encoder_dtype
)
tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")
return {"text_encoder": text_encoder, "tokenizer": tokenizer}
def load_latent_models(self):
torch.manual_seed(0)
vae = AutoencoderKLWan(
base_dim=3,
z_dim=16,
dim_mult=[1, 1, 1, 1],
num_res_blocks=1,
temperal_downsample=[False, True, True],
)
# TODO(aryan): Upload dummy checkpoints to the Hub so that we don't have to do this.
# Doing so overrides things like _keep_in_fp32_modules
vae.to(self.vae_dtype)
self.vae_config = vae.config
return {"vae": vae}
def load_diffusion_models(self):
torch.manual_seed(0)
transformer = WanTransformer3DModel(
patch_size=(1, 2, 2),
num_attention_heads=2,
attention_head_dim=12,
in_channels=16,
out_channels=16,
text_dim=32,
freq_dim=256,
ffn_dim=32,
num_layers=2,
cross_attn_norm=True,
qk_norm="rms_norm_across_heads",
rope_max_seq_len=32,
)
# TODO(aryan): Upload dummy checkpoints to the Hub so that we don't have to do this.
# Doing so overrides things like _keep_in_fp32_modules
transformer.to(self.transformer_dtype)
scheduler = FlowMatchEulerDiscreteScheduler()
return {"transformer": transformer, "scheduler": scheduler}
|