llava.py

# SPDX-License-Identifier: Apache-2.0

from abc import abstractmethod
from collections.abc import Iterable, Mapping, Sequence
from functools import cached_property
from typing import (Final, List, Literal, Optional, Protocol, Set, Tuple,
                    TypedDict, TypeVar, Union, cast)

import torch
import torch.nn as nn
from packaging.version import Version
from transformers import (BatchFeature, CLIPVisionConfig, LlavaConfig,
                          PixtralVisionConfig, PretrainedConfig,
                          SiglipVisionConfig)
from transformers import __version__ as TRANSFORMERS_VERSION
from transformers.models.llava import LlavaProcessor
from transformers.models.pixtral import PixtralProcessor

from vllm.config import VllmConfig
from vllm.inputs import InputProcessingContext
from vllm.model_executor.layers.activation import get_act_fn
from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                               RowParallelLinear)
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                    MultiModalInputs, MultiModalKwargs,
                                    NestedTensors)
from vllm.multimodal.parse import (ImageEmbeddingItems, ImageProcessorItems,
                                   ImageSize, MultiModalDataItems)
from vllm.multimodal.processing import (BaseMultiModalProcessor,
                                        BaseProcessingInfo, ProcessingCache,
                                        PromptReplacement, PromptUpdate)
from vllm.multimodal.profiling import BaseDummyInputsBuilder, ProcessorInputs
from vllm.sequence import IntermediateTensors
from vllm.utils import JSONTree, flatten_2d_lists, json_map_leaves

from .clip import CLIPVisionModel
from .interfaces import SupportsMultiModal, SupportsPP
from .pixtral import (PixtralHFVisionModel,
                      get_pixtral_hf_image_feature_grid_size)
from .siglip import SiglipVisionModel
from .utils import (AutoWeightsLoader, flatten_bn, init_vllm_registered_model,
                    maybe_prefix, merge_multimodal_embeddings)
from .vision import get_vision_encoder_info


class LlavaImagePixelInputs(TypedDict):
    type: Literal["pixel_values"]
    data: Union[torch.Tensor, List[torch.Tensor]]
    """
    Shape: `(batch_size * num_images, num_channels, height, width)`

    Note that `height` or `width` may be different per batch and image,
    in which case the data is passed as a list instead of a batched tensor.
    """

    feat_is_patch: Union[torch.Tensor, List[torch.Tensor]]
    """
    A boolean mask indicating which image features correspond
    to patch tokens.

    Shape: `(batch_size, num_crops, num_patch)`
    """

    embed_is_patch: Union[torch.Tensor, List[torch.Tensor]]
    """
    A boolean mask indicating which image embeddings correspond
    to patch tokens.
    
    Shape: `(batch_size, num_embeds)`
    """

    num_crops: torch.Tensor
    """Shape: `(batch_size, num_images)`"""


class LlavaImageEmbeddingInputs(TypedDict):
    type: Literal["image_embeds"]
    data: torch.Tensor
    """Shape: `(batch_size * num_images, image_feature_size, hidden_size)`

    `hidden_size` must match the hidden size of language model backbone.
    """

    feat_is_patch: Union[torch.Tensor, List[torch.Tensor]]
    """
    A boolean mask indicating which image features correspond
    to patch tokens.

    Shape: `(batch_size, num_crops, num_patch)`
    """

    embed_is_patch: Union[torch.Tensor, List[torch.Tensor]]
    """
    A boolean mask indicating which image embeddings correspond
    to patch tokens.
    
    Shape: `(batch_size, num_embeds)`
    """

    num_crops: torch.Tensor
    """Shape: `(batch_size, num_images)`"""


LlavaImageInputs = Union[LlavaImagePixelInputs, LlavaImageEmbeddingInputs]


class LlavaMultiModalProjector(nn.Module):

    def __init__(self,
                 vision_hidden_size: int,
                 text_hidden_size: int,
                 projector_hidden_act: str,
                 multimodal_projector_bias: bool,
                 quant_config: Optional[QuantizationConfig] = None,
                 prefix: str = ""):
        super().__init__()

        self.linear_1 = ColumnParallelLinear(vision_hidden_size,
                                             text_hidden_size,
                                             bias=multimodal_projector_bias,
                                             quant_config=quant_config,
                                             prefix=f"{prefix}.linear_1")
        self.act = get_act_fn(projector_hidden_act)
        self.linear_2 = RowParallelLinear(text_hidden_size,
                                          text_hidden_size,
                                          bias=multimodal_projector_bias,
                                          quant_config=quant_config,
                                          prefix=f"{prefix}.linear_2")

    def forward(self, image_features: torch.Tensor) -> torch.Tensor:
        hidden_states, _ = self.linear_1(image_features)
        hidden_states = self.act(hidden_states)
        hidden_states, _ = self.linear_2(hidden_states)
        return hidden_states


class LlavaLikeConfig(Protocol):
    vision_config: Final[PretrainedConfig]
    image_token_index: Final[int]
    vision_feature_select_strategy: Final[str]
    vision_feature_layer: Final[Union[int, list[int]]]


class LlavaLikeProcessor(Protocol):
    image_token: Final[str]


class BaseLlavaProcessingInfo(BaseProcessingInfo):

    def get_hf_config(self) -> LlavaLikeConfig:
        return self.ctx.get_hf_config(LlavaConfig)

    def get_vision_encoder_info(self):
        return get_vision_encoder_info(self.get_hf_config())

    @abstractmethod
    def get_hf_processor(self, **kwargs: object) -> LlavaLikeProcessor:
        raise NotImplementedError

    def get_supported_mm_limits(self) -> Mapping[str, Optional[int]]:
        return {"image": None}

    def get_mm_max_tokens_per_item(
        self,
        seq_len: int,
        mm_counts: Mapping[str, int],
    ) -> Mapping[str, int]:
        return {"image": self.get_max_image_tokens()}

    def _apply_feature_select_strategy(
        self,
        strategy: str,
        encoder_num_image_tokens: int,
    ) -> int:
        if strategy == "default":
            return encoder_num_image_tokens - 1
        if strategy == "full":
            return encoder_num_image_tokens

        msg = f"Unexpected feature select strategy: {strategy!r}"
        raise NotImplementedError(msg)

    def get_num_image_tokens(
        self,
        *,
        image_width: int,
        image_height: int,
    ) -> int:
        hf_config = self.get_hf_config()
        vision_encoder_info = self.get_vision_encoder_info()

        return self._apply_feature_select_strategy(
            hf_config.vision_feature_select_strategy,
            vision_encoder_info.get_num_image_tokens(
                image_width=image_width,
                image_height=image_height,
            ),
        )

    def get_image_size_with_most_features(self) -> ImageSize:
        vision_encoder_info = self.get_vision_encoder_info()
        width = height = vision_encoder_info.get_image_size()
        return ImageSize(width=width, height=height)

    def get_max_image_tokens(self) -> int:
        target_width, target_height = self.get_image_size_with_most_features()

        return self.get_num_image_tokens(
            image_width=target_width,
            image_height=target_height,
        )


_I = TypeVar("_I", bound=BaseLlavaProcessingInfo)


class LlavaDummyInputsBuilder(BaseDummyInputsBuilder[_I]):

    def get_dummy_processor_inputs(
        self,
        seq_len: int,
        mm_counts: Mapping[str, int],
    ) -> ProcessorInputs:
        num_images = mm_counts.get("image", 0)

        processor = self.info.get_hf_processor()
        image_token = processor.image_token
        target_width, target_height = \
            self.info.get_image_size_with_most_features()

        mm_data = {
            "image":
            self._get_dummy_images(width=target_width,
                                   height=target_height,
                                   num_images=num_images)
        }

        return ProcessorInputs(
            prompt_text=image_token * num_images,
            mm_data=mm_data,
        )


class LlavaProcessingInfo(BaseLlavaProcessingInfo):

    def get_hf_processor(self, **kwargs: object):
        return self.ctx.get_hf_processor(LlavaProcessor, **kwargs)


class BaseLlavaMultiModalProcessor(BaseMultiModalProcessor[_I]):

    # Copied from BaseMultiModalProcessor
    @abstractmethod
    def _get_mm_fields_config(
        self,
        hf_inputs: BatchFeature,
        hf_processor_mm_kwargs: Mapping[str, object],
    ) -> Mapping[str, MultiModalFieldConfig]:
        raise NotImplementedError

    def _get_prompt_updates(
        self,
        mm_items: MultiModalDataItems,
        hf_processor_mm_kwargs: Mapping[str, object],
        out_mm_kwargs: MultiModalKwargs,
    ) -> Sequence[PromptUpdate]:
        hf_config = self.info.get_hf_config()
        image_token_id = hf_config.image_token_index

        def get_replacement(item_idx: int):
            images = mm_items.get_items(
                "image", (ImageEmbeddingItems, ImageProcessorItems))

            if isinstance(images, ImageEmbeddingItems):
                num_image_tokens = images.get_feature_size(item_idx)
            else:
                image_size = images.get_image_size(item_idx)
                num_image_tokens = self.info.get_num_image_tokens(
                    image_width=image_size.width,
                    image_height=image_size.height,
                )

            return [image_token_id] * num_image_tokens

        return [
            PromptReplacement(
                modality="image",
                target=[image_token_id],
                replacement=get_replacement,
            ),
        ]


class LlavaMultiModalProcessor(
        BaseLlavaMultiModalProcessor[LlavaProcessingInfo]):

    def _get_mm_fields_config(
        self,
        hf_inputs: BatchFeature,
        hf_processor_mm_kwargs: Mapping[str, object],
    ) -> Mapping[str, MultiModalFieldConfig]:
        return dict(
            pixel_values=MultiModalFieldConfig.batched("image"),
            image_embeds=MultiModalFieldConfig.batched("image"),
        )


class PixtralHFProcessingInfo(BaseLlavaProcessingInfo):

    def get_hf_processor(self, **kwargs: object):
        return self.ctx.get_hf_processor(PixtralProcessor, **kwargs)


class PixtralHFMultiModalProcessor(
        BaseMultiModalProcessor[PixtralHFProcessingInfo]):

    def _call_hf_processor(
        self,
        prompt: str,
        mm_data: Mapping[str, object],
        mm_kwargs: Mapping[str, object],
    ) -> BatchFeature:
        processed_outputs = super()._call_hf_processor(
            prompt=prompt,
            mm_data=mm_data,
            mm_kwargs=mm_kwargs,
        )

        pixel_values = processed_outputs.get("pixel_values")
        if pixel_values is not None:
            # Before/after https://github.com/huggingface/transformers/pull/35122
            if Version(TRANSFORMERS_VERSION) <= Version("4.48.3"):
                images = mm_data["images"]
                assert isinstance(images, list)

                # Original output: (1, num_images, C, H, W)
                # New output: (num_images, C, H, W)
                assert (isinstance(pixel_values, list)
                        and len(pixel_values) == 1)
                assert (isinstance(pixel_values[0], list)
                        and len(pixel_values[0]) == len(images))

                processed_outputs["pixel_values"] = pixel_values[0]
            else:
                # Avoid padding since we need the output for each image to be
                # independent of other images for the cache to work correctly
                image_sizes = processed_outputs["image_sizes"]
                assert len(pixel_values) == len(image_sizes)

                processed_outputs["pixel_values"] = [
                    p[:, :h, :w]
                    for p, (h, w) in zip(pixel_values, image_sizes)
                ]

            hf_config = self.info.get_hf_config()

            tile_sizes = [
                get_pixtral_hf_image_feature_grid_size(
                    hf_config.vision_config,
                    image_width=pixel_value.shape[-1],
                    image_height=pixel_value.shape[-2])
                for pixel_value in processed_outputs["pixel_values"]
            ]
            num_crops = torch.tensor([(ncols + 1) * nrows
                                      for ncols, nrows in tile_sizes])
            # Each image may result to masks of different sizes, so we need to
            # flatten the list and later use `num_crops` to get per-image masks.
            embed_is_patch = torch.tensor(
                flatten_2d_lists([([True] * ncols + [False]) * nrows
                                  for ncols, nrows in tile_sizes]))
            processed_outputs["num_crops"] = num_crops
            processed_outputs["embed_is_patch"] = embed_is_patch
            processed_outputs["feat_is_patch"] = embed_is_patch

        return processed_outputs

    def _get_mm_fields_config(
        self,
        hf_inputs: BatchFeature,
        hf_processor_mm_kwargs: Mapping[str, object],
    ) -> Mapping[str, MultiModalFieldConfig]:
        num_crops = hf_inputs.get("num_crops", torch.empty(0)).view(-1)
        return dict(
            feat_is_patch=MultiModalFieldConfig.flat_from_sizes(
                "image", num_crops),
            embed_is_patch=MultiModalFieldConfig.flat_from_sizes(
                "image", num_crops),
            num_crops=MultiModalFieldConfig.batched("image"),
            pixel_values=MultiModalFieldConfig.batched("image"),
            image_embeds=MultiModalFieldConfig.batched("image"),
        )

    def _get_prompt_updates(
        self,
        mm_items: MultiModalDataItems,
        hf_processor_mm_kwargs: Mapping[str, object],
        out_mm_kwargs: MultiModalKwargs,
    ) -> Sequence[PromptUpdate]:
        processor = self.info.get_hf_processor(**hf_processor_mm_kwargs)
        hf_config = self.info.get_hf_config()
        tokenizer = self.info.get_tokenizer()
        vocab = tokenizer.get_vocab()

        image_break_id = vocab[processor.image_break_token]
        image_token_id = hf_config.image_token_index
        image_end_id = vocab[processor.image_end_token]

        vision_config = hf_config.vision_config
        assert isinstance(vision_config, PixtralVisionConfig)

        def get_replacement(item_idx: int):
            images = mm_items.get_items("image", ImageProcessorItems)
            image_size = images.get_image_size(item_idx)

            ncols, nrows = get_pixtral_hf_image_feature_grid_size(
                vision_config,
                image_width=image_size.width,
                image_height=image_size.height,
            )

            tokens = ([image_token_id] * ncols + [image_break_id]) * nrows
            tokens[-1] = image_end_id

            return tokens

        return [
            PromptReplacement(
                modality="image",
                target=[image_token_id],
                replacement=get_replacement,
            ),
        ]


def _build_llava_or_pixtral_hf_info(
    ctx: InputProcessingContext, ) -> BaseLlavaProcessingInfo:
    hf_config = ctx.get_hf_config(LlavaConfig)

    if isinstance(hf_config.vision_config, PixtralVisionConfig):
        return PixtralHFProcessingInfo(ctx)

    return LlavaProcessingInfo(ctx)


def _build_llava_or_pixtral_hf_processor(
    info: _I,
    dummy_inputs: BaseDummyInputsBuilder[_I],
    *,
    cache: Optional[ProcessingCache] = None,
    enable_sanity_checks: bool = True,
) -> BaseMultiModalProcessor:
    if isinstance(info, PixtralHFProcessingInfo):
        return PixtralHFMultiModalProcessor(
            info,
            dummy_inputs,  # type: ignore
            cache=cache,
            enable_sanity_checks=enable_sanity_checks,
        )

    if isinstance(info, LlavaProcessingInfo):
        return LlavaMultiModalProcessor(
            info,
            dummy_inputs,  # type: ignore
            cache=cache,
            enable_sanity_checks=enable_sanity_checks,
        )

    raise NotImplementedError(type(info))


def _get_num_hidden_layers(hf_config: LlavaLikeConfig) -> int:
    """Determine the number of hidden layers to initialize up to in the
    visual encoder.
    
    Args:
        hf_config: Model config with vision feature layer(s).
    """
    feature_layers = hf_config.vision_feature_layer
    num_hidden_layers = hf_config.vision_config.num_hidden_layers
    # If we have one feature layer, initialize up to that layer
    if isinstance(feature_layers, int):
        return _get_layer_index(feature_layers, num_hidden_layers)
    # If we have multiple feature layers, initialize up to the deepest one
    elif isinstance(feature_layers, (list, tuple)):
        return max(
            _get_layer_index(idx, num_hidden_layers) for idx in feature_layers)
    raise TypeError(f"vision_layer_feature type: {type(feature_layers)}"
                    " is not supported")


def _get_layer_index(feature_layer_index: int, num_hidden_layers: int) -> int:
    """Given a signed vision feature layer, get the number of hidden layers
    needed to leverage it.

    Args:
        feature_layer_index: Index of a required layer in the visual encoder.
        num_hidden_layers: The total number of hidden layers in the visual
            encoder.
    """
    if feature_layer_index < 0:
        return num_hidden_layers + feature_layer_index + 1
    return feature_layer_index


def init_vision_tower_for_llava(
    hf_config: LlavaLikeConfig,
    quant_config: Optional[QuantizationConfig],
    *,
    require_post_norm: Optional[bool] = None,
    prefix: str = "",
):
    vision_config = hf_config.vision_config

    # Initialize the vision tower only up to the deepest required feature layer
    num_hidden_layers = _get_num_hidden_layers(hf_config)

    if isinstance(vision_config, CLIPVisionConfig):
        return CLIPVisionModel(
            vision_config,
            quant_config=quant_config,
            num_hidden_layers_override=num_hidden_layers,
            require_post_norm=require_post_norm,
            prefix=prefix,
        )
    elif isinstance(vision_config, SiglipVisionConfig):
        return SiglipVisionModel(
            vision_config,
            quant_config=quant_config,
            num_hidden_layers_override=num_hidden_layers,
            require_post_norm=require_post_norm,
            prefix=prefix,
        )
    elif isinstance(vision_config, PixtralVisionConfig):
        return PixtralHFVisionModel(
            vision_config,
            quant_config=quant_config,
            num_hidden_layers_override=num_hidden_layers,
            require_post_norm=require_post_norm,
            prefix=prefix,
        )

    msg = f"Unsupported vision config: {type(vision_config)}"
    raise NotImplementedError(msg)


@MULTIMODAL_REGISTRY.register_processor(_build_llava_or_pixtral_hf_processor,
                                        info=_build_llava_or_pixtral_hf_info,
                                        dummy_inputs=LlavaDummyInputsBuilder)
class LlavaForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP):

    packed_modules_mapping = {
        "qkv_proj": ["q_proj", "k_proj", "v_proj"],
        "gate_up_proj": ["gate_proj", "up_proj"]
    }

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
        super().__init__()

        config = vllm_config.model_config.hf_config
        quant_config = vllm_config.quant_config
        multimodal_config = vllm_config.model_config.multimodal_config

        self.config = config
        self.multimodal_config = multimodal_config

        # NOTE: These are special cases for Pixtral-12B in the HF-format
        # https://huggingface.co/mistral-community/pixtral-12b/blob/main/config.json  # noqa
        if (config.text_config.architectures is None
                and config.text_config.model_type == "mistral"):
            config.text_config.architectures = ["MistralForCausalLM"]
        if (config.projector_hidden_act is None
                and config.vision_config.hidden_act == "gelu"):
            config.projector_hidden_act = "gelu"

        # TODO: Optionally initializes this for supporting embeddings.
        self.vision_tower = init_vision_tower_for_llava(
            config,
            quant_config,
            require_post_norm=False,
            prefix=maybe_prefix(prefix, "vision_tower"))
        self.multi_modal_projector = LlavaMultiModalProjector(
            vision_hidden_size=config.vision_config.hidden_size,
            text_hidden_size=config.text_config.hidden_size,
            projector_hidden_act=config.projector_hidden_act,
            multimodal_projector_bias=config.multimodal_projector_bias,
            quant_config=quant_config,
            prefix=maybe_prefix(prefix, "multi_modal_projector"))

        self.language_model = init_vllm_registered_model(
            vllm_config=vllm_config,
            hf_config=config.text_config,
            prefix=maybe_prefix(prefix, "language_model"),
        )

        self.make_empty_intermediate_tensors = (
            self.language_model.make_empty_intermediate_tensors)

    @cached_property
    def sampler(self):
        if hasattr(self.language_model, "sampler"):
            return self.language_model.sampler

        return get_sampler()

    def _validate_pixel_values(self, data: torch.Tensor) -> torch.Tensor:
        h = w = self.config.vision_config.image_size
        expected_dims = (3, h, w)
        actual_dims = tuple(data.shape[1:])

        if actual_dims != expected_dims:
            expected_expr = ("batch_size", *map(str, expected_dims))
            raise ValueError(
                f"The expected shape of pixel values is {expected_expr}. "
                f"You supplied {tuple(data.shape)}.")

        return data

    def _parse_and_validate_image_input(
            self, **kwargs: object) -> Optional[LlavaImageInputs]:
        pixel_values = kwargs.pop("pixel_values", None)
        image_embeds = kwargs.pop("image_embeds", None)

        if pixel_values is None and image_embeds is None:
            return None

        feat_is_patch = kwargs.pop("feat_is_patch", None)
        if feat_is_patch is not None and not isinstance(
                feat_is_patch, (torch.Tensor, list)):
            raise ValueError("Incorrect type of feat_is_patch. "
                             f"Got type: {type(feat_is_patch)}")

        embed_is_patch = kwargs.pop("embed_is_patch", None)
        if embed_is_patch is not None and not isinstance(
                embed_is_patch, (torch.Tensor, list)):
            raise ValueError("Incorrect type of embed_is_patch. "
                             f"Got type: {type(embed_is_patch)}")

        num_crops = kwargs.pop("num_crops", None)
        if num_crops is not None and not isinstance(num_crops, torch.Tensor):
            raise ValueError("Incorrect type of num_crops. "
                             f"Got type: {type(num_crops)}")

        if pixel_values is not None:
            if not isinstance(pixel_values, (torch.Tensor, list)):
                raise ValueError("Incorrect type of pixel values. "
                                 f"Got type: {type(pixel_values)}")

            if self.config.vision_config.model_type == "pixtral":
                return LlavaImagePixelInputs(
                    type="pixel_values",
                    data=flatten_bn(pixel_values),
                    feat_is_patch=feat_is_patch,
                    embed_is_patch=embed_is_patch,
                    num_crops=num_crops,
                )

            return LlavaImagePixelInputs(
                type="pixel_values",
                data=self._validate_pixel_values(
                    flatten_bn(pixel_values, concat=True)),
                feat_is_patch=feat_is_patch,
                embed_is_patch=embed_is_patch,
                num_crops=num_crops,
            )

        if image_embeds is not None:
            if not isinstance(image_embeds, (torch.Tensor, list)):
                raise ValueError("Incorrect type of image embeddings. "
                                 f"Got type: {type(image_embeds)}")

            return LlavaImageEmbeddingInputs(
                type="image_embeds",
                data=flatten_bn(image_embeds, concat=True),
                feat_is_patch=feat_is_patch,
                embed_is_patch=embed_is_patch,
                num_crops=num_crops,
            )

        raise AssertionError("This line should be unreachable.")

    def _select_image_features(self, image_features: torch.Tensor, *,
                               strategy: str) -> torch.Tensor:
        # Copied from https://github.com/huggingface/transformers/blob/39c3c0a72af6fbda5614dde02ff236069bb79827/src/transformers/models/llava/modeling_llava.py#L421  # noqa
        if strategy == "default":
            return image_features[:, 1:]
        elif strategy == "full":
            return image_features

        raise ValueError(f"Unexpected select feature strategy: {strategy}")

    def _image_pixels_to_features(
        self,
        vision_tower: Union[CLIPVisionModel, SiglipVisionModel,
                            PixtralHFVisionModel],
        pixel_values: torch.Tensor,
    ) -> torch.Tensor:

        # NOTE: we skip the step to select the vision feature layer since
        # this is already done inside the vision tower
        image_features = vision_tower(pixel_values)

        return self._select_image_features(
            image_features,
            strategy=self.config.vision_feature_select_strategy,
        )

    def _process_image_pixels(self,
                              inputs: LlavaImagePixelInputs) -> torch.Tensor:
        assert self.vision_tower is not None

        pixel_values = inputs["data"]

        return self._image_pixels_to_features(self.vision_tower, pixel_values)

    def _process_image_input(self,
                             image_input: LlavaImageInputs) -> torch.Tensor:

        if image_input["type"] == "image_embeds":
            return image_input["data"]

        assert self.vision_tower is not None
        image_features = self._process_image_pixels(image_input)

        if isinstance(image_features, torch.Tensor):
            return self.multi_modal_projector(image_features)

        feature_sizes = [
            image_feature.shape[0] for image_feature in image_features
        ]

        image_embeds = self.multi_modal_projector(torch.cat(image_features))
        image_embeds = torch.split(image_embeds, feature_sizes)
        return image_embeds

    def _get_mm_embeds(
            self,
            features: torch.Tensor,  # Shape: (num_crop, num_patch, d)
            feat_is_patch: torch.Tensor,  # Shape: (num_crop, num_patch)
            num_crops: torch.Tensor,  # Shape: (num_images,)
            embed_is_patch: torch.Tensor,  # Shape: (num_embeds,)
    ) -> list[torch.Tensor]:
        """Scatter the patch features into a contiguous tensor that corresponds
        to the embedding tokens defined by the multimodal processor.

        Mostly copied from `Molmo._get_mm_embeds`. See following fixme comment.
        """

        # Insert columns of nan values according to `feat_is_patch`. This work
        # ideally should be done in `_process_image_input`, but
        # `_process_image_input` is used in both V0 and V1 path. It's safer to
        # put the logic here.
        # FIXME: Move this logic to `_process_image_input` when v0 is
        # deprecated. Merge this function with `Molmo._get_mm_embeds`.
        feat_is_patch = feat_is_patch.view(-1)
        embed_is_patch = embed_is_patch.view(-1)
        expanded_embedding = torch.full(
            (sum(num_crops), *features.shape[1:]),
            torch.nan,
            dtype=features.dtype).to(features.device)
        expanded_embedding[feat_is_patch] = features

        num_crops_per_image = num_crops.tolist()
        feats_per_image = expanded_embedding.split(num_crops_per_image)
        f_is_patch_per_image = feat_is_patch.split(num_crops_per_image)

        embed_dim = expanded_embedding.shape[-1]
        num_embeds = embed_is_patch.shape[0]

        embeds_in_batch = list[torch.Tensor]()
        for feats, f_is_patch in zip(feats_per_image, f_is_patch_per_image):
            embeds = feats.new_full((num_embeds, embed_dim), torch.nan)
            embeds[embed_is_patch] = feats[f_is_patch]
            embeds_in_batch.append(embeds)

        return embeds_in_batch

    def get_multimodal_embeddings(self, **kwargs) -> Optional[NestedTensors]:
        image_input = self._parse_and_validate_image_input(**kwargs)
        if image_input is None:
            return None
        vision_embeddings = self._process_image_input(image_input)

        if kwargs.get("v0_path", False) or \
            image_input.get("feat_is_patch") is None or \
                image_input.get("embed_is_patch") is None:
            # The path is used for pixtral (V0 only) and llava (V0/V1)
            return vision_embeddings

        nested_emb = [
            self._get_mm_embeds(*args) for args in zip(
                vision_embeddings, image_input["feat_is_patch"],
                image_input["num_crops"], image_input["embed_is_patch"])
        ]
        return flatten_2d_lists(nested_emb)

    def get_input_embeddings(
        self,
        input_ids: torch.Tensor,
        multimodal_embeddings: Optional[NestedTensors] = None,
    ) -> torch.Tensor:
        inputs_embeds = self.language_model.get_input_embeddings(input_ids)
        if multimodal_embeddings is not None:
            # Extract the patch tokens
            patch_embeddings = json_map_leaves(
                lambda x: x[~x.isnan()].view(-1, *x.shape[1:]),
                cast(JSONTree[torch.Tensor], multimodal_embeddings),
            )

            inputs_embeds = merge_multimodal_embeddings(
                input_ids, inputs_embeds, cast(NestedTensors,
                                               patch_embeddings),
                self.config.image_token_index)
        return inputs_embeds

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        intermediate_tensors: Optional[IntermediateTensors] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        **kwargs: object,
    ) -> Union[torch.Tensor, IntermediateTensors]:
        """Run forward pass for LLaVA-1.5.

        One key thing to understand is the `input_ids` already accounts for the
        positions of the to-be-inserted image embeddings.

        Concretely, consider a text prompt:
        `"USER: <image>\\nWhat's the content of the image?\\nASSISTANT:"`.

        Tokenizer outputs:
        `[1, 3148, 1001, 29901, 29871, 32000, 29871, 13, 5618, 29915, 29879,
        278, 2793, 310, 278, 1967, 29973, 13, 22933, 9047, 13566, 29901]`.

        To reserve space in KV cache, we have to insert placeholder tokens
        before they are inputted to the model, so the input processor prepends
        additional image tokens (denoted as `32000`), resulting in:
        `[1, 3148, 1001, 29901, 29871, 32000, ..., 32000, 29871, 13, 5618,
        29915, 29879, 278, 2793, 310, 278, 1967, 29973, 13, 22933, 9047, 13566,
        29901]`.

        We insert 575 tokens so that including the original image token in the
        input, there are a total of 576 (24 * 24) image tokens, which
        corresponds to the number of image tokens inputted to the language
        model, i.e. the number of image tokens outputted by the visual encoder.

        This way, the `positions` and `attn_metadata` are consistent
        with the `input_ids`.

        Args:
            input_ids: Flattened (concatenated) input_ids corresponding to a
                batch.
            pixel_values: The pixels in each input image.

        See also:
            :class:`LlavaImageInputs`
        """
        if intermediate_tensors is not None:
            inputs_embeds = None

        # NOTE: In v1, inputs_embeds is always generated at model runner, this
        # condition is for v0 compatibility.
        elif inputs_embeds is None:
            kwargs.update({"v0_path": True})
            vision_embeddings = self.get_multimodal_embeddings(**kwargs)
            inputs_embeds = self.get_input_embeddings(input_ids,
                                                      vision_embeddings)
            input_ids = None

        hidden_states = self.language_model.model(input_ids,
                                                  positions,
                                                  intermediate_tensors,
                                                  inputs_embeds=inputs_embeds)

        return hidden_states

    def compute_logits(
        self,
        hidden_states: torch.Tensor,
        sampling_metadata: SamplingMetadata,
    ) -> Optional[torch.Tensor]:
        return self.language_model.compute_logits(hidden_states,
                                                  sampling_metadata)

    def sample(
        self,
        logits: torch.Tensor,
        sampling_metadata: SamplingMetadata,
    ) -> Optional[SamplerOutput]:
        return self.language_model.sample(logits, sampling_metadata)

    def load_weights(self, weights: Iterable[Tuple[str,
                                                   torch.Tensor]]) -> Set[str]:
        loader = AutoWeightsLoader(self)
        return loader.load_weights(weights)


class MantisProcessingInfo(LlavaProcessingInfo):

    def get_hf_processor(self, **kwargs: object):
        hf_config = self.get_hf_config()
        vision_info = self.get_vision_encoder_info()

        kwargs.setdefault("patch_size", vision_info.get_patch_size())

        if Version(TRANSFORMERS_VERSION) < Version("4.48"):
            # BUG: num_additional_image_tokens = 0 but treated as 1,
            # so we set vision_feature_select_strategy to None to offset this
            kwargs.setdefault("vision_feature_select_strategy", None)
        else:
            # FIXED: https://github.com/huggingface/transformers/pull/33424/files#diff-6a37acc21efcadaae622b079b2712a131131448ff64262bd219aa346aeec38faL150
            kwargs.setdefault(
                "vision_feature_select_strategy",
                hf_config.vision_feature_select_strategy,
            )

        return self.ctx.get_hf_processor(LlavaProcessor, **kwargs)


class MantisMultiModalProcessor(LlavaMultiModalProcessor):

    def apply(
        self,
        prompt: Union[str, list[int]],
        mm_data: MultiModalDataDict,
        hf_processor_mm_kwargs: Mapping[str, object],
        return_mm_hashes: bool = False,
    ) -> MultiModalInputs:
        hf_config = self.info.get_hf_config()
        image_token_id = hf_config.image_token_index

        # Assume that it doesn't depend on the image size
        num_image_tokens = self.info.get_num_image_tokens(
            image_width=-1,
            image_height=-1,
        )

        result = super().apply(prompt, mm_data, hf_processor_mm_kwargs,
                               return_mm_hashes)

        mm_items = self._to_mm_items(mm_data)
        mm_item_counts = mm_items.get_all_counts()
        mm_kwargs = result["mm_kwargs"]

        # We reimplement the functionality of MLlavaProcessor from
        # https://github.com/TIGER-AI-Lab/Mantis.git
        def get_replacement_mantis(item_idx: int):
            return "".join([
                f"(image {item_idx+1}: <Image>",  # 7 tokens
                "<image>" * num_image_tokens,
                "</Image>)",  # 3 tokens
            ])

        mantis_mm_repls = self._bind_and_group_updates([
            PromptReplacement(
                modality="image",
                target=[image_token_id] * num_image_tokens,
                replacement=get_replacement_mantis,
            )
        ])

        prompt_ids, prompt, _ = self._apply_prompt_updates(
            result["prompt_token_ids"],
            mantis_mm_repls,
            mm_item_counts,
        )

        unbound_orig_repls = self._get_prompt_updates(
            mm_items,
            hf_processor_mm_kwargs,
            mm_kwargs,
        )
        orig_repls = self._bind_and_group_updates(unbound_orig_repls)

        mm_placeholders = self._find_mm_placeholders(
            orig_repls,
            prompt_ids,
            mm_item_counts,
        )
        self._validate_mm_placeholders(mm_placeholders, mm_item_counts)

        mm_placeholder_ranges = {
            modality: [item.to_range() for item in placeholders]
            for modality, placeholders in mm_placeholders.items()
        }

        return MultiModalInputs(
            type="multimodal",
            prompt=prompt,
            prompt_token_ids=prompt_ids,
            mm_kwargs=mm_kwargs,
            mm_placeholders=mm_placeholder_ranges,
        )


# To use this model, please use
# `--hf_overrides '{"architectures": ["MantisForConditionalGeneration"]}'`
@MULTIMODAL_REGISTRY.register_processor(MantisMultiModalProcessor,
                                        info=MantisProcessingInfo,
                                        dummy_inputs=LlavaDummyInputsBuilder)
class MantisForConditionalGeneration(LlavaForConditionalGeneration):
    pass