多模态支持

本文档将引导您完成扩展基本模型以使其接受多模态输入的步骤。

1. 更新基础vLLM模型¶

假设您已经按照这些步骤在vLLM中实现了模型。请按如下方式进一步更新模型：

在 forward 方法中为每个对应于多模态输入的输入张量保留一个关键字参数，如下例所示：

  def forward(
      self,
      input_ids: torch.Tensor,
      positions: torch.Tensor,
+     pixel_values: torch.Tensor,
  ) -> SamplerOutput:

更方便的是，您可以简单地将 **kwargs 传递给 forward 方法，并从中检索多模态输入的关键字参数。

实现 get_multimodal_embeddings 方法，该方法返回通过模型的多模态tokenizer运行多模态输入后获得的embeddings。下面提供了一个典型实现模式的样板，但您可以根据自己的需要进行调整。

class YourModelForImage2Seq(nn.Module):
    ...

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

        assert self.vision_encoder is not None
        image_features = self.vision_encoder(image_input)
        return self.multi_modal_projector(image_features)

    def get_multimodal_embeddings(
            self, **kwargs: object) -> Optional[MultiModalEmbeddings]:

        # Validate the multimodal input keyword arguments
        image_input = self._parse_and_validate_image_input(**kwargs)
        if image_input is None:
            return None

        # Run multimodal inputs through encoder and projector
        vision_embeddings = self._process_image_input(image_input)
        return vision_embeddings

警告

The returned `multimodal_embeddings` must be either a **3D [torch.Tensor][]** of shape `(num_items, feature_size, hidden_size)`, or a **list / tuple of 2D [torch.Tensor][]'s** of shape `(feature_size, hidden_size)`, so that `multimodal_embeddings[i]` retrieves the embeddings generated from the `i`-th multimodal data item (e.g, image) of the request.

实现 [get_input_embeddings][vllm.model_executor.models.interfaces.SupportsMultiModal.get_input_embeddings] 方法，将 multimodal_embeddings 与来自 input_ids 的文本embeddings合并。如果模型的输入处理已正确实现（参见以下部分），则您可以利用我们提供的实用函数轻松合并embeddings。

from .utils import merge_multimodal_embeddings

class YourModelForImage2Seq(nn.Module):
    ...

    def get_input_embeddings(
        self,
        input_ids: torch.Tensor,
        multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
    ) -> torch.Tensor:

        # `get_input_embeddings` should already be implemented for the language 
        # model as one of the requirements of basic vLLM model implementation.
        inputs_embeds = self.language_model.get_input_embeddings(input_ids)

        if multimodal_embeddings is not None:
            inputs_embeds = merge_multimodal_embeddings(
                input_ids=input_ids, 
                inputs_embeds=inputs_embeds, 
                multimodal_embeddings=multimodal_embeddings,
                placeholder_token_id=self.config.image_token_index)

        return inputs_embeds

实现 get_language_model getter方法，以稳定访问底层语言模型。

class YourModelForImage2Seq(nn.Module):
    ...

    def get_language_model(self) -> torch.nn.Module:
        # Change `language_model` according to your implementation.
        return self.language_model

完成上述步骤后，使用 SupportsMultiModal 接口更新模型类。

+ from vllm.model_executor.models.interfaces import SupportsMultiModal

- class YourModelForImage2Seq(nn.Module):
+ class YourModelForImage2Seq(nn.Module, SupportsMultiModal):

注意

模型类不必命名为 *ForCausalLM。请查看 HuggingFace Transformers文档获取一些示例。

2. 指定处理信息¶

接下来，创建一个 BaseProcessingInfo 的子类，以提供与HF处理相关的基本信息。

最大输入项数¶

您需要覆盖抽象方法 get_supported_mm_limits，以返回模型支持的每种模态的最大输入项数。

例如，如果模型支持任意数量的图片，但每个prompt只支持一个视频

def get_supported_mm_limits(self) -> Mapping[str, Optional[int]]:
    return {"image": None, "video": 1}

3. 指定虚拟输入¶

然后，继承 BaseDummyInputsBuilder，以构建用于HF处理和内存分析的虚拟输入。

用于内存分析¶

覆盖抽象方法 get_dummy_text 和 get_dummy_mm_data，以构建用于内存分析的虚拟输入。这些虚拟输入应导致模型的 worst-case 内存使用，以便vLLM为其保留正确的内存量。

假设内存使用量随token数量增加而增加，则可以构建虚拟输入以最大化输出embeddings的数量，该数量与占位符特征token的数量相同。

基本示例：LLaVA无输入占位符：Fuyu

查看HF的 LlavaForConditionalGeneration 代码

# https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/llava/modeling_llava.py#L530-L544
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)
    .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)

每张图片的占位符特征token数量是 image_features.shape[1]。 image_features 在 get_image_features 方法内部计算。

# https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/llava/modeling_llava.py#L290-L300
image_outputs = self.vision_tower(pixel_values, output_hidden_states=True)

selected_image_feature = image_outputs.hidden_states[vision_feature_layer]
if vision_feature_select_strategy == "default":
    selected_image_feature = selected_image_feature[:, 1:]
elif vision_feature_select_strategy == "full":
    selected_image_feature = selected_image_feature
else:
    raise ValueError(f"Unexpected select feature strategy: {self.config.vision_feature_select_strategy}")
image_features = self.multi_modal_projector(selected_image_feature)
return image_features

我们可以推断出 image_features.shape[1] 是基于视觉塔（对于 llava-hf/llava-1.5-7b-hf 模型来说是 CLIPVisionModel）的 image_outputs.hidden_states.shape[1]。此外，我们只需要序列长度（张量的第二个维度）即可获得 image_features.shape[1]。序列长度由 CLIPVisionTransformer 中的初始隐藏状态决定，因为注意力机制不会改变输出隐藏状态的序列长度。

# https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/clip/modeling_clip.py#L1094-L1102
hidden_states = self.embeddings(pixel_values, interpolate_pos_encoding=interpolate_pos_encoding)
hidden_states = self.pre_layrnorm(hidden_states)

encoder_outputs = self.encoder(
    inputs_embeds=hidden_states,
    output_attentions=output_attentions,
    output_hidden_states=output_hidden_states,
    return_dict=return_dict,
)

要找到序列长度，我们转向 CLIPVisionEmbeddings 的代码

# https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/clip/modeling_clip.py#L247-L257
target_dtype = self.patch_embedding.weight.dtype
patch_embeds = self.patch_embedding(pixel_values.to(dtype=target_dtype))  # shape = [*, width, grid, grid]
patch_embeds = patch_embeds.flatten(2).transpose(1, 2)

class_embeds = self.class_embedding.expand(batch_size, 1, -1)
embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
if interpolate_pos_encoding:
    embeddings = embeddings + self.interpolate_pos_encoding(embeddings, height, width)
else:
    embeddings = embeddings + self.position_embedding(self.position_ids)
return embeddings

我们可以推断出 embeddings.shape[1] == self.num_positions，其中

# https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/clip/modeling_clip.py#L195-L196
self.num_patches = (self.image_size // self.patch_size) ** 2
self.num_positions = self.num_patches + 1

总的来说，一张图片的占位符特征token数量可以计算为

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

    image_size = hf_config.vision_config.image_size
    patch_size = hf_config.vision_config.patch_size

    num_image_tokens = (image_size // patch_size) ** 2 + 1
    if hf_processor.vision_feature_select_strategy == "default":
        num_image_tokens -= 1

    return num_image_tokens

注意，图片token的数量不依赖于图片的宽度和高度。我们可以简单地使用一个虚拟的 image_size 来计算多模态性能分析数据。

# NOTE: In actuality, this is usually implemented as part of the
# model's subclass of `BaseProcessingInfo`, but we show it as is
# here for simplicity.
def get_image_size_with_most_features(self) -> ImageSize:
    hf_config = self.get_hf_config()
    width = height = hf_config.image_size
    return ImageSize(width=width, height=height)

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

    target_width, target_height = \
        self.info.get_image_size_with_most_features()

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

对于文本，我们只需从模型配置中扩展多模态图片token，以匹配所需的图片数量。

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

    processor = self.info.get_hf_processor()
    image_token = processor.image_token

    return image_token * num_images

查看HF的 FuyuForCausalLM 代码

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/modeling_fuyu.py#L311-L322
if image_patches is not None and past_key_values is None:
    patch_embeddings = [
        self.vision_embed_tokens(patch.to(self.vision_embed_tokens.weight.dtype))
        .squeeze(0)
        .to(inputs_embeds.device)
        for patch in image_patches
    ]
    inputs_embeds = self.gather_continuous_embeddings(
        word_embeddings=inputs_embeds,
        continuous_embeddings=patch_embeddings,
        image_patch_input_indices=image_patches_indices,
    )

batch中第 i 个item的占位符特征token数量是 patch_embeddings[i].shape[0]，这与 image_patches[i].shape[0] 相同，即 num_total_patches。

与LLaVA不同，Fuyu没有在建模文件中定义patch的数量。我们可以从哪里获取更多信息？考虑到模型输入来自 FuyuProcessor 的输出，让我们查看预处理文件。

图片输出是通过在 FuyuProcessor 内部调用 FuyuImageProcessor.preprocess，然后再调用 FuyuImageProcessor.preprocess_with_tokenizer_info 获得的。

在 FuyuImageProcessor.preprocess 中，图片会被调整大小并填充到目标 FuyuImageProcessor.size，并返回调整大小后（但填充前）的尺寸作为元数据。

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/processing_fuyu.py#L541-L544
image_encoding = self.image_processor.preprocess(images, **output_kwargs["images_kwargs"])
batch_images = image_encoding["images"]
image_unpadded_heights = image_encoding["image_unpadded_heights"]
image_unpadded_widths = image_encoding["image_unpadded_widths"]

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/image_processing_fuyu.py#L480-L
if do_resize:
    batch_images = [
        [self.resize(image, size=size, input_data_format=input_data_format) for image in images]
        for images in batch_images
    ]

image_sizes = [get_image_size(images[0], channel_dim=input_data_format) for images in batch_images]
image_unpadded_heights = [[image_size[0]] for image_size in image_sizes]
image_unpadded_widths = [[image_size[1]] for image_size in image_sizes]

if do_pad:
    batch_images = [
        [
            self.pad_image(
                image,
                size=size,
                mode=padding_mode,
                constant_values=padding_value,
                input_data_format=input_data_format,
            )
            for image in images
        ]
        for images in batch_images
    ]

在 FuyuImageProcessor.preprocess_with_tokenizer_info 中，图片会根据此元数据分割成patches。

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/processing_fuyu.py#L417-L425
model_image_input = self.image_processor.preprocess_with_tokenizer_info(
    image_input=tensor_batch_images,
    image_present=image_present,
    image_unpadded_h=image_unpadded_heights,
    image_unpadded_w=image_unpadded_widths,
    image_placeholder_id=image_placeholder_id,
    image_newline_id=image_newline_id,
    variable_sized=True,
)

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/image_processing_fuyu.py#L638-L658
image_height, image_width = image.shape[1], image.shape[2]
if variable_sized:  # variable_sized=True
    new_h = min(
        image_height,
        math.ceil(image_unpadded_h[batch_index, subseq_index] / patch_height) * patch_height,
    )
    new_w = min(
        image_width,
        math.ceil(image_unpadded_w[batch_index, subseq_index] / patch_width) * patch_width,
    )
    image = image[:, :new_h, :new_w]
    image_height, image_width = new_h, new_w

num_patches = self.get_num_patches(image_height=image_height, image_width=image_width)
tensor_of_image_ids = torch.full(
    [num_patches], image_placeholder_id, dtype=torch.int32, device=image_input.device
)
patches = self.patchify_image(image=image.unsqueeze(0)).squeeze(0)
assert num_patches == patches.shape[0]

patch的数量则由 FuyuImageProcessor.get_num_patches 定义。

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/image_processing_fuyu.py#L552-L562
patch_size = patch_size if patch_size is not None else self.patch_size
patch_height, patch_width = self.patch_size["height"], self.patch_size["width"]

if image_height % patch_height != 0:
    raise ValueError(f"{image_height=} must be divisible by {patch_height}")
if image_width % patch_width != 0:
    raise ValueError(f"{image_width=} must be divisible by {patch_width}")

num_patches_per_dim_h = image_height // patch_height
num_patches_per_dim_w = image_width // patch_width
num_patches = num_patches_per_dim_h * num_patches_per_dim_w

这些图片patches对应于占位符token（|SPEAKER|）。因此，我们只需要最大化图片patches的数量。由于输入图片首先被调整大小以适应 image_processor.size 范围内，我们可以通过输入尺寸等于 image_processor.size 的图片来最大化图片patches的数量。

def get_image_size_with_most_features(self) -> ImageSize:
    image_processor = self.get_image_processor()
    return ImageSize(width=image_processor.size["width"],
                        height=image_processor.size["height"])

Fuyu不期望HF处理器的输入中包含图片占位符，因此无论图片的数量是多少，虚拟prompt文本都为空。

def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str:
    return ""

对于多模态图片性能分析数据，其逻辑与LLaVA非常相似。

def get_dummy_mm_data(
    self,
    seq_len: int,
    mm_counts: Mapping[str, int],
) -> MultiModalDataDict:
    target_width, target_height = \
        self.info.get_image_size_with_most_features()
    num_images = mm_counts.get("image", 0)

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

4. 指定处理细节¶

之后，创建一个 BaseMultiModalProcessor 的子类，以填写HF处理的缺失细节。

信息

多模态数据处理

覆盖 _get_mm_fields_config，以返回HF处理器输出的与输入多模态项相关的张量schema。

基本示例：LLaVA带后处理：Fuyu

CLIPImageProcessor 的输出是一个简单的张量，形状为 (num_images, num_channels, image_height, image_width)。

# https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/clip/image_processing_clip.py#L339-L345
images = [
    to_channel_dimension_format(image, data_format, input_channel_dim=input_data_format)
    for image in all_images
]

data = {"pixel_values": images}
return BatchFeature(data=data, tensor_type=return_tensors)

因此，我们如下覆盖 _get_mm_fields_config：

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"),
    )

注意

我们的实际代码还额外支持预计算的图片embeddings，可以通过 image_embeds 参数传递给模型。

FuyuImageProcessor.preprocess_with_tokenizer_info 的 image_patches 输出会连接batch中每个item所属的图片的patches。

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/image_processing_fuyu.py#L673-L679
        image_input_ids.append(tensor_of_image_ids)
        image_patches.append(patches)
    else:
        image_input_ids.append(torch.tensor([], dtype=torch.int32, device=image_input.device))

batch_image_input_ids.append(image_input_ids)
batch_image_patches.append(image_patches)

因此，FuyuImageProcessor 输出的 image_patches 形状为 (1, num_images, num_patches, patch_width * patch_height * num_channels)。

为了支持像LLaVA中那样使用 [MultiModalFieldConfig.batched][]，我们通过覆盖 [BaseMultiModalProcessor._call_hf_processor][] 来移除额外的batch维度。

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,
    )

    image_patches = processed_outputs.get("image_patches")
    if image_patches is not None:
        images = mm_data["images"]
        assert isinstance(images, list)

        # Original output: (1, num_images, Pn, Px * Py * C)
        # New output: (num_images, Pn, Px * Py * C)
        assert (isinstance(image_patches, list)
                and len(image_patches) == 1)
        assert (isinstance(image_patches[0], torch.Tensor)
                and len(image_patches[0]) == len(images))

        processed_outputs["image_patches"] = image_patches[0]

    return processed_outputs

注意

我们的实际代码对纯文本输入有特殊处理，以防止HF处理器发出不必要的警告。

这使得我们可以如下覆盖 _get_mm_fields_config：

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

Prompt更新¶

覆盖 _get_prompt_updates，以返回一个 PromptUpdate 实例列表。

每个 PromptUpdate 实例指定HF处理器执行的更新操作（例如：插入、替换）。

基本示例：LLaVA处理额外token：Fuyu

查看HF的 LlavaProcessor 代码

# https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/llava/processing_llava.py#L167-L170
prompt_strings = []
for sample in text:
    sample = sample.replace(self.image_token, self.image_token * num_image_tokens)
    prompt_strings.append(sample)

它只是简单地将每个输入的 image_token 重复与占位符特征token数量（num_image_tokens）相等的次数。基于此，我们如下覆盖 _get_prompt_updates：

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", ImageProcessorItems)

        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,
        ),
    ]

回顾步骤2中的特征token布局

|SPEAKER||SPEAKER|...|SPEAKER||NEWLINE|
|SPEAKER||SPEAKER|...|SPEAKER||NEWLINE|
...
|SPEAKER||SPEAKER|...|SPEAKER||NEWLINE|

我们定义一个辅助函数直接返回 ncols 和 nrows。

def get_image_feature_grid_size(
    self,
    *,
    image_width: int,
    image_height: int,
) -> tuple[int, int]:
    image_processor = self.get_image_processor()
    target_width = image_processor.size["width"]
    target_height = image_processor.size["height"]
    patch_width = image_processor.patch_size["width"]
    patch_height = image_processor.patch_size["height"]

    if not (image_width <= target_width and image_height <= target_height):
        height_scale_factor = target_height / image_height
        width_scale_factor = target_width / image_width
        optimal_scale_factor = min(height_scale_factor, width_scale_factor)

        image_height = int(image_height * optimal_scale_factor)
        image_width = int(image_width * optimal_scale_factor)

    ncols = math.ceil(image_width / patch_width)
    nrows = math.ceil(image_height / patch_height)
    return ncols, nrows

基于此，我们最初可以将替换token定义为

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

    ncols, nrows = self.info.get_image_feature_grid_size(
        image_width=image_size.width,
        image_height=image_size.height,
    )

    # `_IMAGE_TOKEN_ID` corresponds to `|SPEAKER|`
    # `_NEWLINE_TOKEN_ID` corresponds to `|NEWLINE|`
    return ([_IMAGE_TOKEN_ID] * ncols + [_NEWLINE_TOKEN_ID]) * nrows

然而，这并不完全正确。在调用 FuyuImageProcessor.preprocess_with_tokenizer_info 后，一个BOS token (<s>) 也会被添加到prompt中。

# https://github.com/huggingface/transformers/blob/v4.48.3/src/transformers/models/fuyu/processing_fuyu.py#L417-L435
model_image_input = self.image_processor.preprocess_with_tokenizer_info(
    image_input=tensor_batch_images,
    image_present=image_present,
    image_unpadded_h=image_unpadded_heights,
    image_unpadded_w=image_unpadded_widths,
    image_placeholder_id=image_placeholder_id,
    image_newline_id=image_newline_id,
    variable_sized=True,
)
prompt_tokens, prompts_length = _tokenize_prompts_with_image_and_batch(
    tokenizer=self.tokenizer,
    prompts=prompts,
    scale_factors=scale_factors,
    max_tokens_to_generate=self.max_tokens_to_generate,
    max_position_embeddings=self.max_position_embeddings,
    add_BOS=True,
    add_beginning_of_answer_token=True,
)

为了仅将视觉embeddings分配给图片token，您可以返回一个 PromptUpdateDetails 实例，而不是字符串。

hf_config = self.info.get_hf_config()
bos_token_id = hf_config.bos_token_id  # `<s>`
assert isinstance(bos_token_id, int)

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

    ncols, nrows = self.info.get_image_feature_grid_size(
        image_width=image_size.width,
        image_height=image_size.height,
    )
    image_tokens = ([_IMAGE_TOKEN_ID] * ncols +
                    [_NEWLINE_TOKEN_ID]) * nrows

    return PromptUpdateDetails.select_token_id(
        image_tokens + [bos_token_id],
        embed_token_id=_IMAGE_TOKEN_ID,
    )

最后，注意到HF处理器会从tokenized的prompt中移除 |ENDOFTEXT| token，我们可以搜索它以在字符串的开头进行替换。

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()
    bos_token_id = hf_config.bos_token_id
    assert isinstance(bos_token_id, int)

    tokenizer = self.info.get_tokenizer()
    eot_token_id = tokenizer.bos_token_id
    assert isinstance(eot_token_id, int)

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

        ncols, nrows = self.info.get_image_feature_grid_size(
            image_width=image_size.width,
            image_height=image_size.height,
        )
        image_tokens = ([_IMAGE_TOKEN_ID] * ncols +
                        [_NEWLINE_TOKEN_ID]) * nrows

        return PromptUpdateDetails.select_token_id(
            image_tokens + [bos_token_id],
            embed_token_id=_IMAGE_TOKEN_ID,
        )

    return [
        PromptReplacement(
            modality="image",
            target=[eot_token_id],
            replacement=get_replacement_fuyu,
        )
    ]

定义了 BaseProcessingInfo (步骤2)、 BaseDummyInputsBuilder (步骤3) 和 BaseMultiModalProcessor (步骤4) 后，用 {meth}MULTIMODAL_REGISTRY.register_processor <vllm.multimodal.registry.MultiModalRegistry.register_processor> 装饰模型类，将它们注册到多模态注册表中。

  from vllm.model_executor.models.interfaces import SupportsMultiModal
+ from vllm.multimodal import MULTIMODAL_REGISTRY

+ @MULTIMODAL_REGISTRY.register_processor(YourMultiModalProcessor,
+                                         info=YourProcessingInfo,
+                                         dummy_inputs=YourDummyInputsBuilder)
  class YourModelForImage2Seq(nn.Module, SupportsMultiModal):

注意事项¶

不替换地插入特征token¶

一些HF处理器直接插入特征token，而不替换原始prompt中的任何内容。在这种情况下，您可以在 _get_prompt_updates 内部使用 PromptInsertion 而不是 PromptReplacement。

示例

BLIP-2（在prompt开头插入）： vllm/model_executor/models/blip2.py
Florence2（在prompt开头插入）： vllm/model_executor/models/florence2.py
Molmo（在 <|endoftext|> token后插入）： vllm/model_executor/models/molmo.py

_get_prompt_updates 假定每次prompt更新的应用对应一个多模态项。如果HF处理器无论有多少多模态项都执行额外处理，您应该覆盖 _apply_hf_processor_tokens_only，以便处理后的token输入与将HF处理器应用于文本输入的结果一致。这是因为根据我们的设计，token输入会绕过HF处理器。

示例

Chameleon（附加 sep_token）： vllm/model_executor/models/chameleon.py
Fuyu（附加 boa_token）： vllm/model_executor/models/fuyu.py
Molmo（应用未在其他地方定义的chat模板）： vllm/model_executor/models/molmo.py

自定义HF处理器¶

有些模型在HF Hub上没有定义HF处理器类。在这种情况下，您可以定义一个具有与HF处理器相同调用签名的自定义HF处理器，并将其传递给 _call_hf_processor。

示例

DeepSeek-VL2： vllm/model_executor/models/deepseek_vl2.py
InternVL： vllm/model_executor/models/internvl.py
Qwen-VL： vllm/model_executor/models/qwen_vl.py