egogpt/model/language_model/egogpt_qwen.py

from typing import List, Optional, Tuple, Union

import torch
import torch.nn as nn
import transformers
from transformers import (
    AutoConfig,
    AutoModelForCausalLM,
    Qwen2Config,
    Qwen2ForCausalLM,
    Qwen2Model,
)
from transformers.generation.utils import GenerateOutput
from transformers.modeling_outputs import CausalLMOutputWithPast

from ..egogpt_arch import EgoGPTMetaForCausalLM, EgoGPTMetaModel


class EgoGPTConfigQwen(Qwen2Config):
    model_type = "egogpt_qwen"


class EgoGPTQwenModel(EgoGPTMetaModel, Qwen2Model):
    config_class = EgoGPTConfigQwen

    def __init__(self, config: Qwen2Config):
        super(EgoGPTQwenModel, self).__init__(config)


class EgoGPTQwenForCausalLM(Qwen2ForCausalLM, EgoGPTMetaForCausalLM):
    config_class = EgoGPTConfigQwen

    def __init__(self, config):
        super(Qwen2ForCausalLM, self).__init__(config)

        config.rope_scaling = None
        self.model = EgoGPTQwenModel(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_model(self):
        return self.model

    def forward(
        self,
        input_ids: torch.LongTensor = 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,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        speech: Optional[torch.FloatTensor] = None,
        speech_lengths: Optional[torch.LongTensor] = None,
        images: Optional[torch.FloatTensor] = None,
        image_sizes: Optional[List[List[int]]] = None,
        modalities: Optional[List[str]] = ["image"],
        return_dict: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
    ) -> Union[Tuple, CausalLMOutputWithPast]:
        if inputs_embeds is None:
            (
                input_ids,
                position_ids,
                attention_mask,
                past_key_values,
                inputs_embeds,
                labels,
            ) = self.prepare_inputs_labels_for_speech_and_text(
                input_ids,
                position_ids,
                attention_mask,
                past_key_values,
                labels,
                speech,
                speech_lengths,
                images,
                image_sizes,
                modalities,
            )

        return super().forward(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            labels=labels,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

    @torch.no_grad()
    def generate(
        self,
        inputs: Optional[torch.Tensor] = None,
        speech: Optional[torch.Tensor] = None,
        speech_lengths: Optional[torch.Tensor] = None,
        images: Optional[torch.FloatTensor] = None,
        image_sizes: Optional[List[List[int]]] = None,
        modalities: Optional[List[str]] = ["image"],
        **kwargs,
    ) -> Union[GenerateOutput, torch.LongTensor]:
        position_ids = kwargs.pop("position_ids", None)
        attention_mask = kwargs.pop("attention_mask", None)
        if "inputs_embeds" in kwargs:
            raise NotImplementedError("`inputs_embeds` is not supported")

        if speech is not None:
            (
                inputs,
                position_ids,
                attention_mask,
                _,
                inputs_embeds,
                _,
            ) = self.prepare_inputs_labels_for_speech_and_text(
                inputs,
                position_ids,
                attention_mask,
                None,
                None,
                speech,
                speech_lengths,
                images,
                image_sizes,
                modalities,
            )
        else:
            inputs_embeds = self.get_model().embed_tokens(inputs)

        return super().generate(
            position_ids=position_ids,
            attention_mask=attention_mask,
            inputs_embeds=inputs_embeds,
            **kwargs,
        )

    def prepare_inputs_for_generation(
        self, input_ids, past_key_values=None, inputs_embeds=None, **kwargs
    ):
        speech = kwargs.pop("speech", None)
        speech_lengths = kwargs.pop("speech_lengths", None)
        inputs = super().prepare_inputs_for_generation(
            input_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            **kwargs,
        )
        if speech is not None:
            inputs["speech"] = speech
            inputs["speech_lengths"] = speech_lengths
        return inputs


AutoConfig.register("egogpt_qwen", EgoGPTConfigQwen)
AutoModelForCausalLM.register(EgoGPTConfigQwen, EgoGPTQwenForCausalLM)