mllama.py

# flake8: noqa
# coding=utf-8
# Copyright 2024 the HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# vllm-project: no copyright
"""PyTorch Mllama model."""

from typing import List, Optional, Tuple, Union

import torch
import torch.utils.checkpoint

from transformers.modeling_outputs import CausalLMOutputWithPast
from transformers.utils import (
    add_start_docstrings,
    logging,
)

# TRACING: imports
from transformers.models.mllama.modeling_mllama import (
    MLLAMA_START_DOCSTRING,
    MllamaForConditionalGeneration,
)

logger = logging.get_logger(__name__)


# TRACING: This function is not traceable
def _prepare_cross_attention_mask(
    cross_attention_mask: torch.Tensor,
    num_vision_tokens: int,
    dtype: str,
) -> Tuple[torch.Tensor, torch.Tensor]:
    # reshape so it can be used by attn module
    # TRACING: cannot unpack cross_attention_mask with arbitrary number of args
    # batch_size, text_total_length, *_ = cross_attention_mask.shape
    batch_size, text_total_length = cross_attention_mask.shape[:2]

    cross_attention_mask = cross_attention_mask.repeat_interleave(num_vision_tokens, dim=3)
    cross_attention_mask = cross_attention_mask.view(batch_size, text_total_length, -1)
    cross_attention_mask = cross_attention_mask.unsqueeze(1)

    # invert the mask
    inverted_cross_attn_mask = (1.0 - cross_attention_mask).to(dtype)
    cross_attention_mask = inverted_cross_attn_mask.masked_fill(
        inverted_cross_attn_mask.to(torch.bool), torch.finfo(dtype).min
    )

    # apply full-row bias, which return 4D tensor of shape [B, H, S1, 1] where value is 0 if the a full row in cross attn mask's
    # last dimension contains negative infinity values, otherwise it's 1
    negative_inf_value = torch.finfo(dtype).min
    full_text_row_masked_out_mask = (
        (cross_attention_mask != negative_inf_value).any(dim=-1).type_as(cross_attention_mask)[..., None]
    )
    cross_attention_mask *= full_text_row_masked_out_mask

    return cross_attention_mask, full_text_row_masked_out_mask


# TRACING: needs to use updated _prepare_cross_attention_mask
@add_start_docstrings(
    """The Mllama model which consists of a vision encoder and a language model.""",
    MLLAMA_START_DOCSTRING,
)
class MllamaForConditionalGeneration(MllamaForConditionalGeneration):
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        pixel_values: Optional[torch.FloatTensor] = None,
        aspect_ratio_mask: Optional[torch.Tensor] = None,
        aspect_ratio_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        cross_attention_mask: Optional[torch.Tensor] = None,
        cross_attention_states: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = 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,
        num_logits_to_keep: int = 0,  # backwards compatibility
        logits_to_keep: Union[int, torch.Tensor] = 0,
    ) -> Union[Tuple, CausalLMOutputWithPast]:
        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

        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 pixel_values is not None and cross_attention_states is not None:
            raise ValueError("`pixel_values` and `cross_attention_states` cannot be provided simultaneously")

        if pixel_values is not None:
            if aspect_ratio_ids is None:
                raise ValueError("`aspect_ratio_ids` must be provided if `pixel_values` is provided")
            # get vision tokens from vision model
            vision_outputs = self.vision_model(
                pixel_values=pixel_values,
                aspect_ratio_ids=aspect_ratio_ids,
                aspect_ratio_mask=aspect_ratio_mask,
                output_hidden_states=output_hidden_states,
                output_attentions=output_attentions,
                return_dict=return_dict,
            )
            cross_attention_states = vision_outputs[0]
            cross_attention_states = self.multi_modal_projector(cross_attention_states).reshape(
                -1, cross_attention_states.shape[-2], self.hidden_size
            )

        if cross_attention_mask is not None:
            # TRACING: use updated function
            cross_attention_mask, full_text_row_masked_out_mask = _prepare_cross_attention_mask(
                cross_attention_mask,
                num_vision_tokens=self.vision_model.num_patches,
                dtype=self.dtype,
            )
        else:
            full_text_row_masked_out_mask = None

        if cross_attention_mask is not None and cache_position is not None:
            cross_attention_mask = cross_attention_mask[:, :, cache_position]
            full_text_row_masked_out_mask = full_text_row_masked_out_mask[:, :, cache_position]

        outputs = self.language_model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            cross_attention_states=cross_attention_states,
            cross_attention_mask=cross_attention_mask,
            full_text_row_masked_out_mask=full_text_row_masked_out_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            inputs_embeds=inputs_embeds,
            labels=labels,
            output_hidden_states=output_hidden_states,
            output_attentions=output_attentions,
            return_dict=return_dict,
            cache_position=cache_position,
            logits_to_keep=logits_to_keep,
        )

        return outputs