Add huggingface llava

src/liger_kernel/transformers/__init__.py

@@ -10,6 +10,7 @@
 from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_gemma  # noqa: F401
 from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_gemma2  # noqa: F401
 from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_llama  # noqa: F401
+from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_llava  # noqa: F401
 from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_mistral  # noqa: F401
 from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_mixtral  # noqa: F401
 from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_mllama  # noqa: F401

src/liger_kernel/transformers/model/llava.py

@@ -0,0 +1,164 @@
+from typing import List
+from typing import Optional
+from typing import Tuple
+from typing import Union
+
+import torch
+import torch.nn as nn
+
+from transformers.models.llava.modeling_llava import _CONFIG_FOR_DOC
+from transformers.models.llava.modeling_llava import LLAVA_INPUTS_DOCSTRING
+from transformers.models.llava.modeling_llava import LlavaCausalLMOutputWithPast
+from transformers.models.llava.modeling_llava import logger
+from transformers.utils import add_start_docstrings_to_model_forward
+from transformers.utils import replace_return_docstrings
+
+from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
+
+
+@add_start_docstrings_to_model_forward(LLAVA_INPUTS_DOCSTRING)
+@replace_return_docstrings(output_type=LlavaCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+def lce_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    pixel_values: torch.FloatTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    vision_feature_layer: Optional[int] = None,
+    vision_feature_select_strategy: Optional[str] = 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,
+    num_logits_to_keep: int = 0,
+) -> Union[Tuple, LlavaCausalLMOutputWithPast]:
+    r"""
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        num_logits_to_keep (`int`, *optional*):
+            Calculate logits for the last `num_logits_to_keep` tokens. If `0`, calculate logits for all
+            `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+            token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from PIL import Image
+    >>> import requests
+    >>> from transformers import AutoProcessor, LlavaForConditionalGeneration
+
+    >>> model = LlavaForConditionalGeneration.from_pretrained("llava-hf/llava-1.5-7b-hf")
+    >>> processor = AutoProcessor.from_pretrained("llava-hf/llava-1.5-7b-hf")
+
+    >>> prompt = "USER: <image>\nWhat's the content of the image? ASSISTANT:"
+    >>> url = "https://www.ilankelman.org/stopsigns/australia.jpg"
+    >>> image = Image.open(requests.get(url, stream=True).raw)
+
+    >>> inputs = processor(images=image, text=prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(**inputs, max_new_tokens=15)
+    >>> processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "USER:  \nWhat's the content of the image? ASSISTANT: The image features a busy city street with a stop sign prominently displayed"
+    ```"""
+    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
+    )
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+    vision_feature_layer = (
+        vision_feature_layer if vision_feature_layer is not None else self.config.vision_feature_layer
+    )
+    vision_feature_select_strategy = (
+        vision_feature_select_strategy
+        if vision_feature_select_strategy is not None
+        else self.config.vision_feature_select_strategy
+    )
+
+    if (input_ids is None) ^ (inputs_embeds is not None):
+        raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+    if pixel_values is not None and inputs_embeds is not None:
+        raise ValueError(
+            "You cannot specify both pixel_values and inputs_embeds at the same time, and must specify either one"
+        )
+
+    if inputs_embeds is None:
+        inputs_embeds = self.get_input_embeddings()(input_ids)
+
+    if pixel_values is not None:
+        image_features = self.get_image_features(
+            pixel_values=pixel_values,
+            vision_feature_layer=vision_feature_layer,
+            vision_feature_select_strategy=vision_feature_select_strategy,
+        )
+
+        n_image_tokens = (input_ids == self.config.image_token_index).sum().item()
+        n_image_features = image_features.shape[0] * image_features.shape[1]
+        if n_image_tokens != n_image_features:
+            raise ValueError(
+                f"Image features and image tokens do not match: tokens: {n_image_tokens}, features {n_image_features}"
+            )
+        special_image_mask = (input_ids == self.config.image_token_index).unsqueeze(-1)
+        special_image_mask = special_image_mask.expand_as(inputs_embeds).to(inputs_embeds.device)
+        image_features = image_features.to(inputs_embeds.device, inputs_embeds.dtype)
+        inputs_embeds = inputs_embeds.masked_scatter(special_image_mask, image_features)
+
+    outputs = self.language_model.model(
+        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,
+        num_logits_to_keep=num_logits_to_keep,
+    )
+    hidden_states = outputs[0]
+
+    loss = None
+    logits = None
+
+    if self.training and (labels is not None):
+        # Shift so that tokens < n predict n
+        if attention_mask is not None:
+            # we use the input attention mask to shift the logits and labels, because it is 2D.
+            # we also crop attn mask in case it is longer, which happens in PrefixTuning with peft
+            shift_attention_mask = attention_mask[:, -(hidden_states.shape[1] - 1) :].to(hidden_states.device)
+            shift_hidden_states = hidden_states[..., :-1, :][
+                shift_attention_mask.to(hidden_states.device) != 0
+            ].contiguous()
+            shift_labels = labels[..., 1:][shift_attention_mask.to(labels.device) != 0].contiguous()
+        else:
+            shift_hidden_states = hidden_states[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+
+        lce = LigerFusedLinearCrossEntropyLoss()
+        loss = lce(self.language_model.lm_head.weight, shift_hidden_states, shift_labels)
+
+    if not return_dict:
+        # NOTE: This part has not been tested.
+        output = (outputs.logits,) + outputs[1:]
+        return (outputs.loss,) + output if outputs.loss is not None else output
+
+    return LlavaCausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+        image_hidden_states=image_features if pixel_values is not None else None,
+    )

src/liger_kernel/transformers/monkey_patch.py

@@ -19,6 +19,7 @@
 from liger_kernel.transformers.model.gemma2 import lce_forward_deprecated as gemma2_lce_forward_deprected
 from liger_kernel.transformers.model.llama import lce_forward as llama_lce_forward
 from liger_kernel.transformers.model.llama import lce_forward_deprecated as llama_lce_forward_deprecated
+from liger_kernel.transformers.model.llava import lce_forward as llava_lce_forward
 from liger_kernel.transformers.model.mistral import lce_forward as mistral_lce_forward
 from liger_kernel.transformers.model.mixtral import lce_forward as mixtral_lce_forward
 from liger_kernel.transformers.model.mixtral import lce_forward_deprecated as mixtral_lce_forward_deprecated
@@ -133,6 +134,56 @@ def apply_liger_kernel_to_llama(
                 _patch_rms_norm_module(decoder_layer.post_attention_layernorm)
 
 
+def apply_liger_kernel_to_llava(
+    cross_entropy: bool = False,
+    fused_linear_cross_entropy: bool = True,
+    model: PreTrainedModel = None,
+    **kwargs,
+) -> None:
+    """
+    Apply Liger kernels to replace original implementation in HuggingFace Llava models.
+    NOTE: Llava is not available in transformers<4.45.0
+
+    Args:
+        rope (bool): Whether to apply Liger's rotary position embedding. Default is True.
+        cross_entropy (bool): Whether to apply Liger's cross entropy loss. Default is False.
+        fused_linear_cross_entropy (bool):
+            Whether to apply Liger's fused linear cross entropy loss. Default is True.
+            `cross_entropy` and `fused_linear_cross_entropy` cannot both be True.
+            If `fused_linear_cross_entropy` is True, the logits will not be materialized but more memory efficient.
+        rms_norm (bool): Whether to apply Liger's RMSNorm. Default is True.
+        swiglu (bool): Whether to apply Liger's SwiGLU MLP. Default is True.
+        model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
+        loaded. Default is None.
+    """
+    assert not (
+        cross_entropy and fused_linear_cross_entropy
+    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+
+    from transformers.models.llava import modeling_llava
+
+    if cross_entropy:
+        logger.warning(TRANSFORMER_DEPRECATION_WARNING)
+        modeling_llava.CrossEntropyLoss = LigerCrossEntropyLoss
+    if fused_linear_cross_entropy:
+        modeling_llava.LlavaForConditionalGeneration.forward = llava_lce_forward
+
+    if model is not None:
+        if model.config.text_config.model_type in MODEL_TYPE_TO_APPLY_LIGER_FN:
+            MODEL_TYPE_TO_APPLY_LIGER_FN[model.config.text_config.model_type](
+                cross_entropy=False,
+                fused_linear_cross_entropy=False,
+                **kwargs,
+            )
+
+        if model.config.vision_config.model_type in MODEL_TYPE_TO_APPLY_LIGER_FN:
+            MODEL_TYPE_TO_APPLY_LIGER_FN[model.config.vision_config.model_type](
+                cross_entropy=False,
+                fused_linear_cross_entropy=False,
+                **kwargs,
+            )
+
+
 def apply_liger_kernel_to_mllama(
     rope: bool = True,
     cross_entropy: bool = False,
@@ -740,6 +791,7 @@ def apply_liger_kernel_to_phi3(
     "gemma": apply_liger_kernel_to_gemma,
     "gemma2": apply_liger_kernel_to_gemma2,
     "llama": apply_liger_kernel_to_llama,
+    "llava": apply_liger_kernel_to_llava,
     "mllama": apply_liger_kernel_to_mllama,
     "mllama_text_model": apply_liger_kernel_to_mllama,
     "mistral": apply_liger_kernel_to_mistral,