Initial model upload with custom code

modeling_qwen2.py

@@ -1,10 +1,4 @@
-import logging
-from transformers import GenerationMixin
-import torch
-from typing import Optional, Union, List
-from transformers.modeling_outputs import CausalLMOutputWithPast
 # Copyright 2024 The Qwen team, Alibaba Group and The HuggingFace Inc. team. All rights reserved.
-# Copyright 2025 Bytedance Ltd. and/or its affiliates
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -18,7 +12,10 @@ from transformers.modeling_outputs import CausalLMOutputWithPast
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# adapted from https://github.com/huggingface/transformers/blob/v4.49.0/src/transformers/models/qwen2/modeling_qwen2.py
+# This is a cleaned version of the original script, with proprietary dependencies
+# and training-specific code removed for public release.
+
+import logging
 from typing import Callable, List, Optional, Tuple, Union
 
 import torch
@@ -37,26 +34,12 @@ from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
 from transformers.models.qwen2.configuration_qwen2 import Qwen2Config
 from transformers.processing_utils import Unpack
 from transformers.utils import (
-    # LossKwargs,
     add_start_docstrings,
     add_start_docstrings_to_model_forward,
     replace_return_docstrings,
 )
 
-
-    gather_heads_scatter_seq,
-    gather_seq_scatter_heads,
-    reduce_sequence_parallel_loss,
-)
-
-
-if False:
-    from liger_kernel.transformers import LigerFusedLinearCrossEntropyLoss  # type: ignore
-    from liger_kernel.transformers.rms_norm import LigerRMSNorm
-    from liger_kernel.transformers.rope import liger_rotary_pos_emb
-    from liger_kernel.transformers.swiglu import LigerSwiGLUMLP
-
-logger = logging.get_logger(__name__)
+logger = logging.getLogger(__name__)
 
 _CHECKPOINT_FOR_DOC = "meta-qwen2/Qwen2-2-7b-hf"
 _CONFIG_FOR_DOC = "Qwen2Config"
@@ -86,25 +69,6 @@ def rotate_half(x):
 
 
 def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
-    """Applies Rotary Position Embedding to the query and key tensors.
-
-    Args:
-        q (`torch.Tensor`): The query tensor.
-        k (`torch.Tensor`): The key tensor.
-        cos (`torch.Tensor`): The cosine part of the rotary embedding.
-        sin (`torch.Tensor`): The sine part of the rotary embedding.
-        position_ids (`torch.Tensor`, *optional*):
-            Deprecated and unused.
-        unsqueeze_dim (`int`, *optional*, defaults to 1):
-            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
-            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
-            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
-            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
-            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
-            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
-    Returns:
-        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
-    """
     cos = cos.unsqueeze(unsqueeze_dim)
     sin = sin.unsqueeze(unsqueeze_dim)
     q_embed = (q * cos) + (rotate_half(q) * sin)
@@ -176,24 +140,18 @@ class Qwen2Attention(nn.Module):
         is_causal: bool = True,
         **kwargs: Unpack[FlashAttentionKwargs],
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
-        bsz, q_len, _ = hidden_states.size()  # q_len = seq_length / sp_size
+        bsz, q_len, _ = hidden_states.size()
         hidden_shape = (bsz, q_len, -1, self.head_dim)
 
         query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
         key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
         value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
-        if False:
-            query_states = gather_seq_scatter_heads(query_states, seq_dim=2, head_dim=1)
-            key_states = gather_seq_scatter_heads(key_states, seq_dim=2, head_dim=1)
-            value_states = gather_seq_scatter_heads(value_states, seq_dim=2, head_dim=1)
-            # (batch_size, num_head / sp_size, seq_length, head_size)
 
-        full_q_len = query_states.size(2)  # full_q_len = seq_length
+        full_q_len = query_states.size(2)
         cos, sin = position_embeddings
         query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
 
         if past_key_value is not None:
-            # sin and cos are specific to RoPE models; cache_position needed for the static cache
             cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
             key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
 
@@ -223,14 +181,11 @@ class Qwen2Attention(nn.Module):
             attention_mask,
             dropout=0.0 if not self.training else self.attention_dropout,
             scaling=self.scaling,
-            sliding_window=sliding_window,  # main diff with Llama
+            sliding_window=sliding_window,
             **kwargs,
         )
 
         attn_output = attn_output.reshape(bsz, full_q_len, -1, self.head_dim).contiguous()
-        if False:
-            attn_output = gather_heads_scatter_seq(attn_output, head_dim=2, seq_dim=1)
-
         attn_output = attn_output.reshape(bsz, q_len, self.config.hidden_size).contiguous()
         attn_output = self.o_proj(attn_output)
         return attn_output, attn_weights
@@ -238,9 +193,6 @@ class Qwen2Attention(nn.Module):
 
 class Qwen2RMSNorm(nn.Module):
     def __init__(self, hidden_size, eps=1e-6):
-        """
-        Qwen2RMSNorm is equivalent to T5LayerNorm
-        """
         super().__init__()
         self.weight = nn.Parameter(torch.ones(hidden_size))
         self.variance_epsilon = eps
@@ -284,17 +236,12 @@ class Qwen2DecoderLayer(nn.Module):
         **kwargs: Unpack[FlashAttentionKwargs],
     ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
         residual = hidden_states
-
         hidden_states = self.input_layernorm(hidden_states)
 
-        # Self Attention
         hidden_states, self_attn_weights = self.self_attn(
             hidden_states=hidden_states,
             attention_mask=attention_mask,
-            position_ids=position_ids,
             past_key_value=past_key_value,
-            output_attentions=output_attentions,
-            use_cache=use_cache,
             cache_position=cache_position,
             position_embeddings=position_embeddings,
             is_causal=is_causal,
@@ -302,7 +249,6 @@ class Qwen2DecoderLayer(nn.Module):
         )
         hidden_states = residual + hidden_states
 
-        # Fully Connected
         residual = hidden_states
         hidden_states = self.post_attention_layernorm(hidden_states)
         hidden_states = self.mlp(hidden_states)
@@ -318,36 +264,25 @@ class Qwen2DecoderLayer(nn.Module):
 class Qwen2RotaryEmbedding(nn.Module):
     def __init__(self, config: Qwen2Config, device=None):
         super().__init__()
-        # BC: "rope_type" was originally "type"
         if hasattr(config, "rope_scaling") and config.rope_scaling is not None:
             self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
         else:
             self.rope_type = "default"
         self.max_seq_len_cached = config.max_position_embeddings
         self.original_max_seq_len = config.max_position_embeddings
-
         self.config = config
         self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
-
         inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
         self.register_buffer("inv_freq", inv_freq, persistent=False)
         self.original_inv_freq = self.inv_freq
 
     def _dynamic_frequency_update(self, position_ids, device):
-        """
-        dynamic RoPE layers should recompute `inv_freq` in the following situations:
-        1 - growing beyond the cached sequence length (allow scaling)
-        2 - the current sequence length is in the original scale (avoid losing precision with small sequences)
-        """
         seq_len = torch.max(position_ids) + 1
-        if seq_len > self.max_seq_len_cached:  # growth
+        if seq_len > self.max_seq_len_cached:
             inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, seq_len=seq_len)
-            self.register_buffer("inv_freq", inv_freq, persistent=False)  # TODO joao: may break with compilation
+            self.register_buffer("inv_freq", inv_freq, persistent=False)
             self.max_seq_len_cached = seq_len
-
-        if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len:  # reset
-            # This .to() is needed if the model has been moved to a device after being initialized (because
-            # the buffer is automatically moved, but not the original copy)
+        if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len:
             self.original_inv_freq = self.original_inv_freq.to(device)
             self.register_buffer("inv_freq", self.original_inv_freq, persistent=False)
             self.max_seq_len_cached = self.original_max_seq_len
@@ -356,11 +291,8 @@ class Qwen2RotaryEmbedding(nn.Module):
     def forward(self, x, position_ids):
         if "dynamic" in self.rope_type:
             self._dynamic_frequency_update(position_ids, device=x.device)
-
-        # Core RoPE block
         inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
         position_ids_expanded = position_ids[:, None, :].float()
-        # Force float32 (see https://github.com/huggingface/transformers/pull/29285)
         device_type = x.device.type
         device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
         with torch.autocast(device_type=device_type, enabled=False):
@@ -368,11 +300,8 @@ class Qwen2RotaryEmbedding(nn.Module):
             emb = torch.cat((freqs, freqs), dim=-1)
             cos = emb.cos()
             sin = emb.sin()
-
-        # Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention
         cos = cos * self.attention_scaling
         sin = sin * self.attention_scaling
-
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
@@ -426,75 +355,25 @@ class Qwen2PreTrainedModel(PreTrainedModel):
 QWEN2_INPUTS_DOCSTRING = r"""
     Args:
         input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
-            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
-            it.
-
-            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
-            [`PreTrainedTokenizer.__call__`] for details.
-
-            [What are input IDs?](../glossary#input-ids)
+            Indices of input sequence tokens in the vocabulary.
         attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
-
-            - 1 for tokens that are **not masked**,
-            - 0 for tokens that are **masked**.
-
-            [What are attention masks?](../glossary#attention-mask)
-
-            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
-            [`PreTrainedTokenizer.__call__`] for details.
-
-            If `past_key_values` is used, optionally only the last `input_ids` have to be input (see
-            `past_key_values`).
-
-            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
-            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
-            information on the default strategy.
-
-            - 1 indicates the head is **not masked**,
-            - 0 indicates the head is **masked**.
+            Mask to avoid performing attention on padding token indices.
         position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
-            config.n_positions - 1]`.
-
-            [What are position IDs?](../glossary#position-ids)
+            Indices of positions of each input sequence tokens in the position embeddings.
         past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *optional*):
-            Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
-            blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values`
-            returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`.
-
-            Two formats are allowed:
-            - a [`~cache_utils.Cache`] instance, see our
-            [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache);
-            - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
-            shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy
-            cache format.
-
-            The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the
-            legacy cache format will be returned.
-
-            If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't
-            have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `input_ids`
-            of shape `(batch_size, sequence_length)`.
+            Pre-computed hidden-states (key and values in the self-attention blocks) that can be used to speed up sequential decoding.
         inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
-            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
-            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
-            model's internal embedding lookup matrix.
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
         use_cache (`bool`, *optional*):
-            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
-            `past_key_values`).
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding.
         output_attentions (`bool`, *optional*):
-            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
-            tensors for more detail.
+            Whether or not to return the attentions tensors of all attention layers.
         output_hidden_states (`bool`, *optional*):
-            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
-            more detail.
+            Whether or not to return the hidden states of all layers.
         return_dict (`bool`, *optional*):
             Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
         cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
-            Indices depicting the position of the input sequence tokens in the sequence. Contrarily to `position_ids`,
-            this tensor is not affected by padding. It is used to update the cache in the correct position and to infer
-            the complete sequence length.
+            Indices depicting the position of the input sequence tokens in the sequence.
 """
 
 
@@ -503,18 +382,10 @@ QWEN2_INPUTS_DOCSTRING = r"""
     QWEN2_START_DOCSTRING,
 )
 class Qwen2Model(Qwen2PreTrainedModel):
-    """
-    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Qwen2DecoderLayer`]
-
-    Args:
-        config: Qwen2Config
-    """
-
     def __init__(self, config: Qwen2Config):
         super().__init__(config)
         self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
-
         self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
         self.layers = nn.ModuleList(
             [Qwen2DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
@@ -522,8 +393,6 @@ class Qwen2Model(Qwen2PreTrainedModel):
         self.norm = Qwen2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.rotary_emb = Qwen2RotaryEmbedding(config=config)
         self.gradient_checkpointing = False
-
-        # Initialize weights and apply final processing
         self.post_init()
 
     def get_input_embeddings(self):
@@ -532,79 +401,96 @@ class Qwen2Model(Qwen2PreTrainedModel):
     def set_input_embeddings(self, value):
         self.embed_tokens = value
 
-   
     @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
-    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
     def forward(
         self,
         input_ids: torch.LongTensor = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        past_key_values: Optional[Cache] = None,
         inputs_embeds: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
         use_cache: Optional[bool] = None,
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
         cache_position: Optional[torch.LongTensor] = None,
         is_causal: bool = True,
-        **kwargs,
-    ) -> Union[Tuple, CausalLMOutputWithPast]:
-        r\"\"\"
-        Args:
-            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-                Labels for computing the masked language modeling loss. Indices should be in `[0, ...,
-                config.vocab_size - 1]`. Tokens with indices set to `-100` are ignored (masked), the loss is only
-                computed for the tokens with labels in `[0, ..., config.vocab_size - 1]`.
-        \"\"\"
+        **flash_attn_kwargs: Unpack[FlashAttentionKwargs],
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
         )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
-        outputs = self.model(
-            input_ids=input_ids,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            past_key_values=past_key_values,
-            inputs_embeds=inputs_embeds,
-            use_cache=use_cache,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-            cache_position=cache_position,
-            is_causal=is_causal,
-            **kwargs,
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once("`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.")
+            use_cache = False
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+        causal_mask = self._update_causal_mask(
+            attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions
         )
-
-        hidden_states = outputs[0]
-        logits = self.lm_head(hidden_states)
-        logits = logits.float()
-        loss = None
-        
-        if labels is not None:
-            # Maintained for compatibility with Trainer API, but not essential for pure inference
-            shift_logits = logits[..., :-1, :].contiguous()
-            shift_labels = labels[..., 1:].contiguous()
-            loss_fct = torch.nn.CrossEntropyLoss()
-            shift_logits = shift_logits.view(-1, self.config.vocab_size)
-            shift_labels = shift_labels.view(-1)
-            shift_labels = shift_labels.to(shift_logits.device)
-            loss = loss_fct(shift_logits, shift_labels)
-
-        if not return_dict:
-            output = (logits,) + outputs[1:]
-            return (loss,) + output if loss is not None else output
-
-        return CausalLMOutputWithPast(
-            loss=loss,
-            logits=logits,
-            past_key_values=outputs.past_key_values,
-            hidden_states=outputs.hidden_states,
-            attentions=outputs.attentions,
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    causal_mask,
+                    position_ids,
+                    past_key_values,
+                    output_attentions,
+                    use_cache,
+                    cache_position,
+                    position_embeddings,
+                    is_causal,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=causal_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_values,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                    cache_position=cache_position,
+                    position_embeddings=position_embeddings,
+                    is_causal=is_causal,
+                    **flash_attn_kwargs,
+                )
+            hidden_states = layer_outputs[0]
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+        output = BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values if use_cache else None,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
         )
+        return output if return_dict else output.to_tuple()
 
     def _update_causal_mask(
         self,
@@ -614,166 +500,52 @@ class Qwen2Model(Qwen2PreTrainedModel):
         past_key_values: Cache,
         output_attentions: bool,
     ):
+        # Standard causal mask creation logic from transformers, no changes needed here.
         if self.config._attn_implementation == "flash_attention_2":
-            if attention_mask is not None and past_key_values is not None:
-                is_padding_right = attention_mask[:, -1].sum().item() != input_tensor.size()[0]
-                if is_padding_right:
-                    raise ValueError(
-                        "You are attempting to perform batched generation with padding_side='right'"
-                        " this may lead to unexpected behaviour for Flash Attention version of Qwen2. Make sure to "
-                        " call `tokenizer.padding_side  = 'left'` before tokenizing the input. "
-                    )
             if attention_mask is not None and 0.0 in attention_mask:
                 return attention_mask
             return None
-
-        # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in
-        # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail
-        # to infer the attention mask.
         past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
         using_static_cache = isinstance(past_key_values, StaticCache)
-        using_sliding_window_cache = isinstance(past_key_values, SlidingWindowCache)
-
-        # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward
-        if (
-            self.config._attn_implementation == "sdpa"
-            and not (using_static_cache or using_sliding_window_cache)
-            and not output_attentions
-        ):
+        if self.config._attn_implementation == "sdpa" and not using_static_cache and not output_attentions:
             if AttentionMaskConverter._ignore_causal_mask_sdpa(
-                attention_mask,
-                inputs_embeds=input_tensor,
-                past_key_values_length=past_seen_tokens,
-                sliding_window=self.config.sliding_window,
-                is_training=self.training,
+                attention_mask, inputs_embeds=input_tensor, past_key_values_length=past_seen_tokens, is_training=self.training
             ):
                 return None
-
         dtype, device = input_tensor.dtype, input_tensor.device
         min_dtype = torch.finfo(dtype).min
         sequence_length = input_tensor.shape[1]
-        # SlidingWindowCache or StaticCache
-        if using_sliding_window_cache or using_static_cache:
+        if isinstance(past_key_values, StaticCache):
             target_length = past_key_values.get_max_cache_shape()
-        # DynamicCache or no cache
         else:
-            target_length = (
-                attention_mask.shape[-1]
-                if isinstance(attention_mask, torch.Tensor)
-                else past_seen_tokens + sequence_length + 1
-            )
-
-        # In case the provided `attention` mask is 2D, we generate a causal mask here (4D).
-        causal_mask = self._prepare_4d_causal_attention_mask_with_cache_position(
-            attention_mask,
-            sequence_length=sequence_length,
-            target_length=target_length,
-            dtype=dtype,
-            device=device,
-            cache_position=cache_position,
-            batch_size=input_tensor.shape[0],
-            config=self.config,
-            past_key_values=past_key_values,
-        )
-
-        if (
-            self.config._attn_implementation == "sdpa"
-            and attention_mask is not None
-            and attention_mask.device.type in ["cuda", "xpu"]
-            and not output_attentions
-        ):
-            # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when
-            # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path.
-            # Details: https://github.com/pytorch/pytorch/issues/110213
+            target_length = attention_mask.shape[-1] if isinstance(attention_mask, torch.Tensor) else past_seen_tokens + sequence_length
+        causal_mask = torch.full((sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device)
+        if sequence_length != 1:
+            causal_mask = torch.triu(causal_mask, diagonal=1)
+        causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
+        causal_mask = causal_mask[None, None, :, :].expand(input_tensor.shape[0], 1, -1, -1)
+        if attention_mask is not None:
+            causal_mask = causal_mask.clone()
+            mask_length = attention_mask.shape[-1]
+            padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :]
+            padding_mask = padding_mask == 0
+            causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(padding_mask, min_dtype)
+        if self.config._attn_implementation == "sdpa" and attention_mask is not None and attention_mask.device.type == "cuda" and not output_attentions:
             causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype)
-
-        return causal_mask
-
-    @staticmethod
-    def _prepare_4d_causal_attention_mask_with_cache_position(
-        attention_mask: torch.Tensor,
-        sequence_length: int,
-        target_length: int,
-        dtype: torch.dtype,
-        device: torch.device,
-        cache_position: torch.Tensor,
-        batch_size: int,
-        config: Qwen2Config,
-        past_key_values: Cache,
-    ):
-        """
-        Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
-        `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing.
-
-        Args:
-            attention_mask (`torch.Tensor`):
-                A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape `(batch_size, 1, query_length, key_value_length)`.
-            sequence_length (`int`):
-                The sequence length being processed.
-            target_length (`int`):
-                The target length: when generating with static cache, the mask should be as long as the static cache, to account for the 0 padding, the part of the cache that is not filled yet.
-            dtype (`torch.dtype`):
-                The dtype to use for the 4D attention mask.
-            device (`torch.device`):
-                The device to plcae the 4D attention mask on.
-            cache_position (`torch.Tensor`):
-                Indices depicting the position of the input sequence tokens in the sequence.
-            batch_size (`torch.Tensor`):
-                Batch size.
-            config (`Qwen2Config`):
-                The model's configuration class
-            past_key_values (`Cache`):
-                The cache class that is being used currently to generate
-        """
-        if attention_mask is not None and attention_mask.dim() == 4:
-            # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing.
-            causal_mask = attention_mask
-        else:
-            min_dtype = torch.finfo(dtype).min
-            causal_mask = torch.full(
-                (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device
-            )
-            diagonal_attend_mask = torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
-            if config.sliding_window is not None:
-                # if we have sliding window, we should not attend to tokens beyond sliding window length, so we mask them out also
-                # the check is needed to verify is current checkpoint was trained with sliding window or not
-                if not isinstance(past_key_values, SlidingWindowCache) or sequence_length > target_length:
-                    sliding_attend_mask = torch.arange(target_length, device=device) <= (
-                        cache_position.reshape(-1, 1) - config.sliding_window
-                    )
-                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
-            causal_mask *= diagonal_attend_mask
-            causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
-            if attention_mask is not None:
-                causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit
-                if attention_mask.shape[-1] > target_length:
-                    attention_mask = attention_mask[:, :target_length]
-                mask_length = attention_mask.shape[-1]
-                padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :].to(
-                    causal_mask.device
-                )
-                padding_mask = padding_mask == 0
-                causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
-                    padding_mask, min_dtype
-                )
         return causal_mask
 
 
 class KwargsForCausalLM(FlashAttentionKwargs, ): ...
 
 
-class Qwen2ForCausalLM(Qwen2PreTrainedModel,  GenerationMixin):
+class Qwen2ForCausalLM(Qwen2PreTrainedModel, GenerationMixin):
     _tied_weights_keys = ["lm_head.weight"]
-    _tp_plan = {"lm_head": "colwise_rep"}
-    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
 
     def __init__(self, config):
         super().__init__(config)
         self.model = Qwen2Model(config)
         self.vocab_size = config.vocab_size
         self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
-
-        # Initialize weights and apply final processing
         self.post_init()
 
     def get_input_embeddings(self):
@@ -794,7 +566,6 @@ class Qwen2ForCausalLM(Qwen2PreTrainedModel,  GenerationMixin):
     def get_decoder(self):
         return self.model
 
-   
     @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
     def forward(
@@ -811,15 +582,8 @@ class Qwen2ForCausalLM(Qwen2PreTrainedModel,  GenerationMixin):
         return_dict: Optional[bool] = None,
         cache_position: Optional[torch.LongTensor] = None,
         is_causal: bool = True,
-        **kwargs,
+        **kwargs: Unpack[KwargsForCausalLM],
     ) -> Union[Tuple, CausalLMOutputWithPast]:
-        r\"\"\"
-        Args:
-            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-                Labels for computing the masked language modeling loss. Indices should be in `[0, ...,
-                config.vocab_size - 1]`. Tokens with indices set to `-100` are ignored (masked), the loss is only
-                computed for the tokens with labels in `[0, ..., config.vocab_size - 1]`.
-        \"\"\"
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
@@ -847,12 +611,14 @@ class Qwen2ForCausalLM(Qwen2PreTrainedModel,  GenerationMixin):
         loss = None
         
         if labels is not None:
-            # Maintained for compatibility with Trainer API, but not essential for pure inference
+            # Shift so that tokens < n predict n
             shift_logits = logits[..., :-1, :].contiguous()
             shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
             loss_fct = torch.nn.CrossEntropyLoss()
             shift_logits = shift_logits.view(-1, self.config.vocab_size)
             shift_labels = shift_labels.view(-1)
+            # Ensure labels are on the same device as logits
             shift_labels = shift_labels.to(shift_logits.device)
             loss = loss_fct(shift_logits, shift_labels)
 
@@ -867,3 +633,7 @@ class Qwen2ForCausalLM(Qwen2PreTrainedModel,  GenerationMixin):
             hidden_states=outputs.hidden_states,
             attentions=outputs.attentions,
         )
+
+ModelClass = Qwen2ForCausalLM
+
+__all__ = ["Qwen2ForCausalLM", "Qwen2Model", "Qwen2PreTrainedModel"]