modeling_ovis.py

# Copyright (C) 2025 AIDC-AI
#
# 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.

import logging
import os
import importlib.metadata

from packaging import version
from importlib import import_module
from typing import List, Callable, Union, Optional, Dict

import PIL.Image
import torch
from torch import Tensor
from torch.nn import init
from torch.nn.functional import softmax, gumbel_softmax, pad
from transformers.utils import is_flash_attn_2_available
from transformers import PreTrainedModel, AutoModel, AutoTokenizer, AutoModelForCausalLM, AutoImageProcessor
from transformers.generation.utils import GenerateOutput

from .configuration_ovis import BaseVisualTokenizerConfig, Aimv2VisualTokenizerConfig
from .configuration_ovis import OvisConfig, ConversationFormatter
from .configuration_ovis import IGNORE_ID, IMAGE_ATOM_ID, IMAGE_INDICATOR_IDS, IMAGE_TOKEN_ID

# ----------------------------------------------------------------------
#                            Visual Tokenizer
# ----------------------------------------------------------------------
class BaseVisualTokenizer(PreTrainedModel):
    base_model_prefix = "backbone"
    main_input_name = None
    _image_processor_class = None
    _image_processor_kwargs = {}
    _backbone_class = None
    _backbone_name_or_path = None

    def __init__(self, config: BaseVisualTokenizerConfig, *inputs, **kwargs):
        super().__init__(config, *inputs, **kwargs)
        self.image_processor = AutoImageProcessor.from_pretrained(kwargs['image_processor_name_or_path'])
        self.backbone = AutoModel.from_config(self.config.backbone_config)
        head_dim = self.config.vocab_size - len(IMAGE_INDICATOR_IDS)  # reserved tokens for IMAGE_INDICATORS
        self.head = torch.nn.Sequential(
            torch.nn.Linear(
                self.backbone.config.hidden_size * self.config.hidden_stride * self.config.hidden_stride, head_dim,
                bias=False
            ),
            torch.nn.LayerNorm(head_dim)
        )

        assert all((self.image_processor.do_resize,
                    not getattr(self.image_processor, 'do_center_crop', False),
                    self.image_processor.do_rescale,
                    self.image_processor.do_normalize
                    )), f"image_processor `{self.image_processor}` is not supported currently"

    def get_backbone(self):
        return self.backbone

    def get_image_processor(self):
        return self.image_processor

    def mock_input(self):
        height, width = self.get_image_size()
        return torch.zeros(1, 3, height, width), self.construct_image_placeholders((1, 1))

    def get_head(self):
        return self.head

    def get_image_size(self):
        raise NotImplementedError

    @staticmethod
    def construct_image_placeholders(grid):
        image_placeholders = [IMAGE_INDICATOR_IDS[0], IMAGE_ATOM_ID, IMAGE_INDICATOR_IDS[1]]
        if grid[0] * grid[1] > 1:
            for r in range(grid[0]):
                for c in range(grid[1]):
                    image_placeholders.append(IMAGE_ATOM_ID)
                    if c < grid[1] - 1:
                        image_placeholders.append(IMAGE_INDICATOR_IDS[2])
                if r < grid[0] - 1:
                    image_placeholders.append(IMAGE_INDICATOR_IDS[3])
        image_placeholders.append(IMAGE_INDICATOR_IDS[4])
        return image_placeholders

    def preprocess_image(self, image: PIL.Image.Image, max_partition=9, covering_threshold=0.9, convert_to_rgb=True):
        def _preprocess(img: PIL.Image.Image, side):
            # first resize and preprocess
            w, h = img.size
            if w == h:
                new_width = new_height = side
            elif w > h:
                new_width = side
                new_height = int(h / w * new_width)
            else:
                new_height = side
                new_width = int(w / h * new_height)
            new_size = dict(height=new_height, width=new_width)
            pixel_values = self.image_processor.preprocess(img, size=new_size, return_tensors='pt')['pixel_values']

            # then pad to square
            square_values = torch.zeros([1, 3, side, side], dtype=pixel_values.dtype, device=pixel_values.device)
            new_height, new_width = pixel_values.shape[2:]
            if new_height == new_width:
                square_values[:, :, :, :] = pixel_values
            elif new_height > new_width:
                from_index = (side - new_width) // 2
                square_values[:, :, :, from_index:from_index + new_width] = pixel_values
            else:
                from_index = (side - new_height) // 2
                square_values[:, :, from_index:from_index + new_height, :] = pixel_values

            return square_values

        def _partition(img, grid):
            w, h = img.size
            row_height = h // grid[0]
            col_width = w // grid[1]

            partition = []
            for row in range(grid[0]):
                for col in range(grid[1]):
                    left = col * col_width
                    upper = row * row_height
                    right = w if col == grid[1] - 1 else (col + 1) * col_width
                    lower = h if row == grid[0] - 1 else (row + 1) * row_height
                    partition.append((left, upper, right, lower))

            return partition

        def _covering_area(left, upper, right, lower, side):
            w = right - left
            h = lower - upper
            w, h = max(w, h), min(w, h)
            if w > side:
                h = h / w * side
                w = side
            return w * h

        def _get_best_grid(img, side):
            img_area = img.size[0] * img.size[1]

            candidate_grids = []
            for i in range(1, max_partition + 1):
                for j in range(1, max_partition + 1):
                    if i * j <= max_partition:
                        candidate_grids.append((i, j))

            all_grids = []
            good_grids = []
            for grid in candidate_grids:
                partition = _partition(img, grid)
                covering_ratio = sum([_covering_area(*p, side) for p in partition]) / img_area
                assert covering_ratio <= 1.0
                all_grids.append((grid, covering_ratio))
                if covering_ratio > covering_threshold:
                    good_grids.append((grid, covering_ratio))

            if len(good_grids) > 0:
                # pick the good partition with minimum #sub_images and break the tie using covering_ratio
                return sorted(good_grids, key=lambda x: (x[0][0] * x[0][1], -x[1]))[0][0]
            else:
                # pick the partition with maximum covering_ratio and break the tie using #sub_images
                return sorted(all_grids, key=lambda x: (-x[1], x[0][0] * x[0][1]))[0][0]

        if convert_to_rgb and image.mode != 'RGB':
            image = image.convert('RGB')

        sides = self.get_image_size()
        if sides[0] != sides[1]:
            raise ValueError('get_image_size() returns non-square size')
        side = sides[0]
        grid = _get_best_grid(image, side)
        partition = _partition(image, grid)
        crops = [image.crop(p) for p in partition]
        if len(crops) > 1:
            crops.insert(0, image)
        pixel_values = torch.cat([_preprocess(crop, side) for crop in crops], dim=0)
        image_placeholders = self.construct_image_placeholders(grid)
        return pixel_values, image_placeholders

    def tokenize(self, logits):
        def st_argmax(y_soft, dim):  # straight-through softmax
            index = y_soft.max(dim, keepdim=True)[1]
            y_hard = torch.zeros_like(y_soft, memory_format=torch.legacy_contiguous_format).scatter_(dim, index, 1.0)
            ret = y_hard - y_soft.detach() + y_soft
            return ret

        if self.config.tokenize_function == 'softmax':
            tokens = softmax(logits, dim=-1)
        elif self.config.tokenize_function == 'gumbel_argmax':
            tokens = gumbel_softmax(logits, tau=self.config.tau, hard=True)
        elif self.config.tokenize_function == 'st_argmax':
            tokens = st_argmax(logits, dim=-1)
        else:
            raise ValueError(
                f'Invalid `max_type`, expected softmax or gumbel_argmax or st_argmax, but got {self.config.tokenize_function}')
        return tokens

    def encode(self, pixel_values):
        output = self.backbone(pixel_values, output_hidden_states=True, return_dict=True)
        features = output.hidden_states[-1]
        if self.config.drop_cls_token:
            features = features[:, 1:, :]

        # merge number of `hidden_stride * hidden_stride` hidden states together to reduce token sequence length
        # e.g., for hidden_stride=2, this leads to a token length reduction: 1024 -> 256 for aimv2
        if self.config.hidden_stride > 1:
            n, l, d = features.shape  # this `d` maybe different from the above `d
            sqrt_l = int(l ** 0.5)
            assert sqrt_l ** 2 == l, "The token sequence length should be a perfect square."
            features = features.reshape(n, sqrt_l, sqrt_l, d)
            pl = (self.config.hidden_stride - (sqrt_l % self.config.hidden_stride)) % self.config.hidden_stride
            features = pad(features, (0, 0, 0, pl, 0, pl), "constant", 0)
            sqrt_l += pl
            features = features.reshape(n, sqrt_l // self.config.hidden_stride, self.config.hidden_stride,
                                        sqrt_l // self.config.hidden_stride, self.config.hidden_stride, d)
            features = features.permute(0, 1, 3, 2, 4, 5)  # [n, sqrt_l/hs, sqrt_l/hs, hs, hs, d]
            features = features.flatten(3)  # [n, sqrt_l/hs, sqrt_l/hs, hs*hs*d]
            features = features.reshape(
                n, -1, self.config.hidden_stride * self.config.hidden_stride * d)

        return features

    def forward(self, pixel_values) -> torch.Tensor:  # [BatchSize, ImageShape] -> [BatchSize, #Token, VocabSize]
        features = self.encode(pixel_values)
        logits = self.head(features)
        tokens = self.tokenize(logits)
        # tokens' shape is [BatchSize, #Token, VocabSize-5], so padding with [BatchSize, #Token, 5], after
        # which, tokens' shape should become [BatchSize, #Token, VocabSize]
        batch_size, token_len, _ = tokens.shape
        padding_tensor = torch.zeros(size=(batch_size, token_len, len(IMAGE_INDICATOR_IDS)),
                                     dtype=tokens.dtype,
                                     device=tokens.device,
                                     layout=tokens.layout,
                                     requires_grad=False)
        tokens = torch.cat((tokens, padding_tensor), dim=2)
        return tokens


class Aimv2VisualTokenizer(BaseVisualTokenizer):
    config_class = Aimv2VisualTokenizerConfig
    supports_gradient_checkpointing = True
    _no_split_modules = ["AIMv2ViTPreprocessor", "AIMv2Block"]
    _image_processor_kwargs = dict(do_center_crop=False)

    def get_image_size(self):
        height = self.image_processor.crop_size["height"]
        width = self.image_processor.crop_size["width"]
        return height, width


AutoModel.register(Aimv2VisualTokenizerConfig, Aimv2VisualTokenizer)


# ----------------------------------------------------------------------
#                                  Ovis
# ----------------------------------------------------------------------
class VisualEmbedding(torch.nn.Embedding):
    def forward(self, visual_tokens: Tensor) -> Tensor:
        if visual_tokens.dtype in [torch.int8, torch.int16, torch.int32, torch.int64, torch.long]:
            return super().forward(visual_tokens)
        return torch.matmul(visual_tokens, self.weight)

    def reset_parameters(self, mean=0., std=1.) -> None:
        init.normal_(self.weight, mean=mean, std=std)
        self._fill_padding_idx_with_zero()


class OvisPreTrainedModel(PreTrainedModel):
    config_class = OvisConfig
    base_model_prefix = "ovis"


class Ovis(OvisPreTrainedModel):

    def __init__(self, config: OvisConfig, *inputs, **kwargs):
        super().__init__(config, *inputs, **kwargs)
        attn_kwargs = dict()
        if self.config.llm_attn_implementation:
            if self.config.llm_attn_implementation == "flash_attention_2":
                assert (is_flash_attn_2_available() and
                        version.parse(importlib.metadata.version("flash_attn")) >= version.parse("2.6.3")), \
                    "Using `flash_attention_2` requires having `flash_attn>=2.6.3` installed."
            attn_kwargs["attn_implementation"] = self.config.llm_attn_implementation
        self.llm = AutoModelForCausalLM.from_config(self.config.llm_config, **attn_kwargs)
        assert self.config.hidden_size == self.llm.config.hidden_size, "hidden size mismatch"
        self.text_tokenizer = AutoTokenizer.from_pretrained(self.config.name_or_path)
        self.visual_tokenizer = AutoModel.from_config(self.config.visual_tokenizer_config,
                                                      image_processor_name_or_path=self.config.name_or_path)
        self.vte = VisualEmbedding(
            self.config.visual_tokenizer_config.vocab_size,
            self.config.hidden_size,
            device=self.visual_tokenizer.device,
            dtype=self.visual_tokenizer.dtype
        )

        def _merge_modules(modules_list: tuple):
            merged_modules = []
            for modules in modules_list:
                merged_modules.extend(modules if modules else [])
            return merged_modules

        self._no_split_modules = _merge_modules((self.llm._no_split_modules, self.visual_tokenizer._no_split_modules))
        self._skip_keys_device_placement = self.llm._skip_keys_device_placement
        self._keep_in_fp32_modules = _merge_modules(
            (self.llm._keep_in_fp32_modules, self.visual_tokenizer._keep_in_fp32_modules))
        self.is_parallelizable = all((self.llm.is_parallelizable, self.visual_tokenizer.is_parallelizable))
        self.supports_gradient_checkpointing = True
        self._supports_flash_attn_2 = True

    def get_text_tokenizer(self):
        return self.text_tokenizer

    def get_visual_tokenizer(self):
        return self.visual_tokenizer

    def tie_weights(self):
        if not self.config.disable_tie_weight:
            self.get_llm().tie_weights()

    def get_llm(self):
        return self.llm

    def get_vte(self):
        return self.vte

    def get_wte(self):
        return self.llm.get_input_embeddings()

    def get_conversation_formatter(self) -> ConversationFormatter:
        if getattr(self, 'conversation_formatter', None) is None:
            self.conversation_formatter = getattr(import_module(".configuration_ovis", __package__),
                                                  self.config.conversation_formatter_class)(self.text_tokenizer)
        return self.conversation_formatter

    def forward(
        self,
        input_ids: torch.Tensor,
        attention_mask: torch.Tensor,
        labels: Optional[torch.Tensor],
        pixel_values: List[Optional[torch.Tensor]],
        **kwargs
    ):
        # assert self.training, "`forward` can only be used in training. For inference, use `generate`."
        _, inputs_embeds, labels, attention_mask = self.merge_multimodal(
            text_input_ids=input_ids,
            text_attention_masks=attention_mask,
            text_labels=labels,
            pixel_values=pixel_values
        )
        return self.llm(inputs_embeds=inputs_embeds, labels=labels, attention_mask=attention_mask, **kwargs)

    def merge_multimodal(
        self,
        text_input_ids: torch.Tensor,
        text_attention_masks: torch.Tensor,
        text_labels: Optional[torch.Tensor],
        pixel_values: List[Optional[torch.Tensor]],
        left_padding: bool = False
    ):
        input_device = text_input_ids.device
        visual_vocab_szie = self.get_visual_tokenizer().config.vocab_size
        visual_indicator_embeds = self.get_vte()(
            torch.tensor(
                list(range(visual_vocab_szie - 5, visual_vocab_szie)),
                dtype=torch.long,
                device=self.get_visual_tokenizer().device
            )
        ).to(device=input_device)

        if self.training:
            # When training, to be compatible with deepspeed zero, each sample has to include pixel_value tensor.
            # For text-only sample, one can simply use a full zero tensor as pixel_value, which will be ignored
            # (see below in this function); so, the gradient will not be affected.
            num_images = [x.shape[0] for x in pixel_values]
            visual_tokens = self.visual_tokenizer(torch.cat([x for x in pixel_values], dim=0))
            visual_embeds = torch.split(self.get_vte()(visual_tokens).to(dtype=self.dtype, device=input_device),
                                        split_size_or_sections=num_images, dim=0)
            visual_input_ids = torch.split(torch.argmax(visual_tokens, dim=-1).to(device=input_device),
                                           split_size_or_sections=num_images, dim=0)
            visual_labels = [torch.full(x.shape, IGNORE_ID, dtype=torch.long, device=input_device) for x in
                             visual_input_ids]
        else:
            # When inference, sample can include only text with `None` pixel_value
            num_images = [x.shape[0] if x is not None else 0 for x in pixel_values]
            if sum(num_images) > 0:
                visual_tokens = self.visual_tokenizer(torch.cat([x for x in pixel_values if x is not None], dim=0))
                visual_embeds = torch.split(self.get_vte()(visual_tokens).to(dtype=self.dtype, device=input_device),
                                            split_size_or_sections=num_images, dim=0)
                visual_input_ids = torch.split(torch.argmax(visual_tokens, dim=-1).to(device=input_device),
                                               split_size_or_sections=num_images, dim=0)
                visual_labels = [torch.full(x.shape, IGNORE_ID, dtype=torch.long, device=input_device) for x in
                                 visual_input_ids]
            else:
                # just placeholders
                visual_embeds = [None] * len(num_images)
                visual_input_ids = [None] * len(num_images)
                visual_labels = [None] * len(num_images)
            # just placeholders
            if text_labels is None:
                text_labels = torch.full(text_input_ids.shape, IGNORE_ID, dtype=torch.long, device=input_device)

        input_embeds = []
        attention_masks = []
        labels = []
        for text_input_id, text_label, text_attention_mask, visual_embed, visual_input_id, visual_label in zip(
                text_input_ids, text_labels, text_attention_masks, visual_embeds, visual_input_ids, visual_labels
        ):
            placeholder_token_mask = torch.lt(text_input_id, 0)
            text_embed = self.get_wte()(torch.masked_fill(text_input_id, placeholder_token_mask, 0))
            for i, indicator_id in enumerate(IMAGE_INDICATOR_IDS):
                text_embed[text_input_id == indicator_id] = visual_indicator_embeds[i]
            image_atom_positions = torch.where(torch.eq(text_input_id, IMAGE_ATOM_ID))[0].tolist()
            if len(image_atom_positions) > 0:
                input_embed_parts = []
                attention_mask_parts = []
                label_parts = []
                prev_image_atom_position = -1
                for index, image_atom_position in enumerate(image_atom_positions):
                    input_embed_parts.append(
                        text_embed[prev_image_atom_position + 1:image_atom_position, :])
                    label_parts.append(
                        text_label[prev_image_atom_position + 1:image_atom_position])
                    attention_mask_parts.append(
                        text_attention_mask[prev_image_atom_position + 1:image_atom_position])
                    input_embed_parts.append(visual_embed[index])
                    attention_mask_parts.append(
                        torch.ones_like(visual_label[index], dtype=torch.bool))
                    label_parts.append(visual_label[index])
                    prev_image_atom_position = image_atom_position
                if prev_image_atom_position + 1 < text_input_id.shape[0]:
                    input_embed_parts.append(
                        text_embed[prev_image_atom_position + 1:, :])
                    attention_mask_parts.append(
                        text_attention_mask[prev_image_atom_position + 1:])
                    label_parts.append(
                        text_label[prev_image_atom_position + 1:])
                input_embed = torch.cat(input_embed_parts, dim=0)
                attention_mask = torch.cat(attention_mask_parts, dim=0)
                label = torch.cat(label_parts, dim=0)
            else:
                input_embed = text_embed
                attention_mask = text_attention_mask
                label = text_label
                if self.training:
                    # Make visual_embed & visual_indicator_embeds involved in the backward graph,
                    # to be compatible with deepspeed zero and ddp.
                    input_embed += torch.sum(visual_embed * 0.0) + torch.sum(visual_indicator_embeds * 0.0)
            input_embeds.append(input_embed)
            attention_masks.append(attention_mask)
            labels.append(label)

        if self.training:  # padding to self.config.multimodal_max_length for increased training speed
            padding_size = max(0, self.config.multimodal_max_length - len(input_embeds[0]))
            input_embeds[0] = torch.nn.ConstantPad2d((0, 0, 0, padding_size), 0.0)(input_embeds[0])
            attention_masks[0] = torch.nn.ConstantPad1d((0, padding_size), False)(attention_masks[0])
            labels[0] = torch.nn.ConstantPad1d((0, padding_size), IGNORE_ID)(labels[0])
        batch_input_embeds = self.pad_truncate_sequence(input_embeds, batch_first=True, padding_value=0.0, left_padding=left_padding)
        batch_attention_mask = self.pad_truncate_sequence(attention_masks, batch_first=True, padding_value=False, left_padding=left_padding)
        batch_labels = self.pad_truncate_sequence(labels, batch_first=True, padding_value=IGNORE_ID, left_padding=left_padding)

        return visual_input_ids, batch_input_embeds, batch_labels, batch_attention_mask

    def pad_truncate_sequence(self, sequences: List[torch.Tensor], batch_first: bool = True, padding_value: float = 0.0, left_padding: bool = False) -> torch.Tensor:
        if not left_padding:
            pad_sequence = torch.nn.utils.rnn.pad_sequence(sequences, batch_first=batch_first, padding_value=padding_value)
            return pad_sequence[:,:self.config.multimodal_max_length]
        else:
            pad_sequence = torch.nn.utils.rnn.pad_sequence([i.flip(dims=[0]) for i in sequences],batch_first=True, padding_value=padding_value).flip(dims=[1])
            return pad_sequence[:,-self.config.multimodal_max_length:]

    def preprocess_inputs(
        self,
        text_or_conversations: Union[List[Dict], str],
        images: Optional[List[PIL.Image.Image]],
        max_partition=9,
        generation_preface='',
        return_labels=False,
        propagate_exception=True,
        frame_selector=None,
        frame_selector_kwargs=None
    ):
        # convert text to conversations
        if isinstance(text_or_conversations, str):
            conversations = [{
                "from": "human",
                "value": text_or_conversations
            }]
        elif isinstance(text_or_conversations, list):
            conversations = text_or_conversations
        else:
            raise ValueError(f'Invalid type of `text_or_conversations`, expected `List[Dict]` or `str`,'
                             f' but got {type(text_or_conversations)}')

        if frame_selector is not None:
            frame_selector_kwargs = frame_selector_kwargs or {}
            conversations, images = frame_selector(conversations=conversations, frames=images, **frame_selector_kwargs)

        # format conversations
        prompt, raw_input_ids, raw_labels = self.get_conversation_formatter().format(
            conversations, generation_preface=generation_preface)

        # place image placeholders
        input_ids = []
        labels = []
        pixel_values = []
        invalidate_label = False
        image_token_indices = [i for i, v in enumerate(raw_input_ids) if v == IMAGE_TOKEN_ID]
        last_image_token_index = -1
        for i in range(len(image_token_indices)):
            head = 0 if i == 0 else image_token_indices[i - 1] + 1
            tail = image_token_indices[i]
            last_image_token_index = tail
            input_ids.extend(raw_input_ids[head:tail])
            labels.extend(raw_labels[head:tail])
            try:
                image = images[i]
                raw_pixel_values, image_placeholders = self.visual_tokenizer.preprocess_image(
                    image, max_partition=max_partition)
            except Exception as e:
                if propagate_exception:
                    raise e
                logging.exception(e)
                invalidate_label = True
                raw_pixel_values, image_placeholders = self.visual_tokenizer.mock_input()
            input_ids.extend(image_placeholders)
            labels.extend([IGNORE_ID] * len(image_placeholders))
            pixel_values.append(raw_pixel_values)
        input_ids.extend(raw_input_ids[last_image_token_index + 1:])
        labels.extend(raw_labels[last_image_token_index + 1:])

        # return tensors
        input_ids = torch.tensor(input_ids, dtype=torch.long)
        labels = torch.tensor([IGNORE_ID] * len(labels) if invalidate_label else labels, dtype=torch.long)
        pixel_values = torch.cat(pixel_values, dim=0) if len(pixel_values) > 0 else None

        if return_labels:
            return prompt, input_ids, pixel_values, labels
        else:
            return prompt, input_ids, pixel_values

    def save_pretrained(
        self,
        save_directory: Union[str, os.PathLike],
        is_main_process: bool = True,
        state_dict: Optional[dict] = None,
        save_function: Callable = torch.save,
        push_to_hub: bool = False,
        max_shard_size: Union[int, str] = "5GB",
        safe_serialization: bool = True,
        variant: Optional[str] = None,
        token: Optional[Union[str, bool]] = None,
        save_peft_format: bool = True,
        **kwargs
    ):
        super().save_pretrained(save_directory,
                                is_main_process=is_main_process,
                                state_dict=state_dict,
                                save_function=save_function,
                                safe_serialization=safe_serialization)
        self.get_text_tokenizer().save_pretrained(save_directory)
        self.get_visual_tokenizer().get_image_processor().save_pretrained(save_directory)

    def generate(
        self,
        inputs: Optional[torch.Tensor] = None,
        **kwargs
    ) -> Union[GenerateOutput, torch.LongTensor]:
        _, inputs_embeds, labels, attention_mask = self.merge_multimodal(
            text_input_ids=inputs,
            text_attention_masks=kwargs.pop('attention_mask'),
            text_labels=None,
            pixel_values=kwargs.pop('pixel_values'),
            left_padding=True
        )
        inputs_embeds = inputs_embeds.detach()
        torch.cuda.empty_cache()

        return self.llm.generate(inputs=None, inputs_embeds=inputs_embeds, attention_mask=attention_mask, **kwargs)