multimodal.md 36.9 KB
Newer Older
1
# Multi-Modal Support
2

3
This document walks you through the steps to extend a basic model so that it accepts [multi-modal inputs](../../features/multimodal_inputs.md).
4
5
6

## 1. Update the base vLLM model

7
It is assumed that you have already implemented the model in vLLM according to [these steps](basic.md).
8
9
Further update the model as follows:

10
11
- Implement [get_placeholder_str][vllm.model_executor.models.interfaces.SupportsMultiModal.get_placeholder_str] to define the placeholder string which is used to represent the multi-modal item in the text prompt. This should be consistent with the chat template of the model.

12
    ??? code
13
14
15
16
17
18

        ```python
        class YourModelForImage2Seq(nn.Module):
            ...

            @classmethod
19
            def get_placeholder_str(cls, modality: str, i: int) -> str | None:
20
21
22
23
24
25
                if modality.startswith("image"):
                    return "<image>"

                raise ValueError("Only image modality is supported")
        ```

26
- Inside `__init__` method, initialize the language components of the model inside [_mark_language_model][vllm.model_executor.models.interfaces.SupportsMultiModal._mark_language_model], and the multimodal components of the model inside [_mark_tower_model][vllm.model_executor.models.interfaces.SupportsMultiModal._mark_tower_model], e.g.:
27

28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
    ```python
        def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
            super().__init__()

            config = vllm_config.model_config.hf_config

            with self._mark_tower_model(vllm_config, "image"):
                self.vision_encoder = ...
                self.multi_modal_projector = ...

            with self._mark_language_model(vllm_config):
                self.language_model = init_vllm_registered_model(
                    vllm_config=vllm_config,
                    hf_config=config.text_config,
                    prefix=maybe_prefix(prefix, "language_model"),
                )
    ```
45

zhuwenwen's avatar
zhuwenwen committed
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
- Remove the embedding part from the [forward][torch.nn.Module.forward] method:
    - Move the multi-modal embedding to [embed_multimodal][vllm.model_executor.models.interfaces.SupportsMultiModal.embed_multimodal].
    - The text embedding and embedding merge are handled automatically by a default implementation of [embed_input_ids][vllm.model_executor.models.interfaces.SupportsMultiModal.embed_input_ids]. It does not need to be overridden in most cases.

    ```diff
      def forward(
          self,
          input_ids: torch.Tensor | None,
    -     pixel_values: torch.Tensor,
          positions: torch.Tensor,
          intermediate_tensors: IntermediateTensors | None = None,
          inputs_embeds: torch.Tensor | None = None,
      ) -> torch.Tensor:
    -     if inputs_embeds is None:
    -         inputs_embeds = self.get_input_embeddings()(input_ids)
    -
    -     if pixel_values is not None:
    -         image_features = self.get_image_features(
    -             pixel_values=pixel_values,
    -         )
    -         special_image_mask = self.get_placeholder_mask(
    -             input_ids,
    -             inputs_embeds=inputs_embeds,
    -             image_features=image_features,
    -         )
    -         inputs_embeds = inputs_embeds.masked_scatter(
    -             special_image_mask,
    -             image_features,
    -         )

           hidden_states = self.language_model(
               input_ids,
               positions,
               intermediate_tensors,
               inputs_embeds=inputs_embeds,
           )
         ...
  
    +  def embed_multimodal(
    +      self,
    +      pixel_values: torch.Tensor,
    +  ) -> MultiModalEmbeddings | None:
    +      return self.get_image_features(
    +          pixel_values=pixel_values,
    +      )
    ```
92

zhuwenwen's avatar
zhuwenwen committed
93
    Below we provide a boilerplate of a typical implementation pattern of [embed_multimodal][vllm.model_executor.models.interfaces.SupportsMultiModal.embed_multimodal], but feel free to adjust it to your own needs.
94

zhuwenwen's avatar
zhuwenwen committed
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
    ```python
    def _process_image_input(self, image_input: YourModelImageInputs) -> torch.Tensor:
        image_features = self.vision_encoder(image_input)
        return self.multi_modal_projector(image_features)

    def embed_multimodal(
        self,
        **kwargs: object,
    ) -> MultiModalEmbeddings | None:
        # 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
    ```
113

114
115
!!! important
    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.
116

117
118
119
!!! note
    By default, vLLM merges the multimodal embeddings into text embeddings depending on the information of their locations defined in
    [PlaceholderRange][vllm.multimodal.inputs.PlaceholderRange] from input processing.
120
    This logic can be found at [embed_input_ids][vllm.model_executor.models.interfaces.SupportsMultiModal.embed_input_ids].
121

122
    You may override this method if additional logic is required for your model when merging embeddings.
123
124
125
126
127
128
129
130
131
132
133

- Once the above steps are done, update the model class with the [SupportsMultiModal][vllm.model_executor.models.interfaces.SupportsMultiModal] interface.

  ```diff
  + from vllm.model_executor.models.interfaces import SupportsMultiModal

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

!!! note
134
135
    The model class does not have to be named `*ForCausalLM`.
    Check out [the HuggingFace Transformers documentation](https://huggingface.co/docs/transformers/model_doc/auto#multimodal) for some examples.
136
137
138
139
140
141
142
143
144
145
146
147
148
149

## 2. Specify processing information

Next, create a subclass of [BaseProcessingInfo][vllm.multimodal.processing.BaseProcessingInfo]
to provide basic information related to HF processing.

### Maximum number of input items

You need to override the abstract method [get_supported_mm_limits][vllm.multimodal.processing.BaseProcessingInfo.get_supported_mm_limits]
to return the maximum number of input items for each modality supported by the model.

For example, if the model supports any number of images but only one video per prompt:

```python
150
def get_supported_mm_limits(self) -> Mapping[str, int | None]:
151
152
153
154
155
    return {"image": None, "video": 1}
```

## 3. Specify dummy inputs

156
157
Then, inherit [BaseDummyInputsBuilder][vllm.multimodal.processing.BaseDummyInputsBuilder] to construct dummy inputs for
HF processing. The processed outputs are also used for memory profiling.
158

159
Override the abstract methods [get_dummy_text][vllm.multimodal.processing.BaseDummyInputsBuilder.get_dummy_text] and [get_dummy_mm_data][vllm.multimodal.processing.BaseDummyInputsBuilder.get_dummy_mm_data] to construct dummy inputs. These dummy inputs should result in the worst-case memory usage of the model so that vLLM can reserve the correct amount of memory for it.
160
161
162
163
164
165
166

Assuming that the memory usage increases with the number of tokens, the dummy inputs can be constructed to maximize the number of output embeddings, which is the same number as placeholder feature tokens.

=== "Basic example: LLaVA"

    Looking at the code of HF's `LlavaForConditionalGeneration`:

167
    ??? code
168

169
170
171
172
173
174
175
176
177
178
179
180
181
182
        ```python
        # 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)
183
        )
184
185
186
        image_features = image_features.to(inputs_embeds.device, inputs_embeds.dtype)
        inputs_embeds = inputs_embeds.masked_scatter(special_image_mask, image_features)
        ```
187
188
189
190

    The number of placeholder feature tokens per image is `image_features.shape[1]`.
    `image_features` is calculated inside the `get_image_features` method:

191
    ??? code
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206

        ```python
        # 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
        ```
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228

    We can infer that `image_features.shape[1]` is based on `image_outputs.hidden_states.shape[1]` from the vision tower
    (`CLIPVisionModel` for the [`llava-hf/llava-1.5-7b-hf`](https://huggingface.co/llava-hf/llava-1.5-7b-hf) model).
    Moreover, we only need the sequence length (the second dimension of the tensor) to get `image_features.shape[1]`.
    The sequence length is determined by the initial hidden states in `CLIPVisionTransformer` since the attention
    mechanism doesn't change the sequence length of the output hidden states.

    ```python
    # 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,
    )
    ```

    To find the sequence length, we turn to the code of `CLIPVisionEmbeddings`:

229
    ??? code
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244

        ```python
        # 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
        ```
245
246
247
248
249
250
251
252
253
254
255

    We can infer that `embeddings.shape[1] == self.num_positions`, where

    ```python
    # 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
    ```

    Overall, the number of placeholder feature tokens for an image can be calculated as:

256
    ??? code
257

258
259
260
261
262
263
264
265
266
        ```python
        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()
267

268
269
            image_size = hf_config.vision_config.image_size
            patch_size = hf_config.vision_config.patch_size
270

271
272
273
274
275
276
            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
        ```
277
278
279
280

    Notice that the number of image tokens doesn't depend on the image width and height.
    We can simply use a dummy `image_size` to calculate the multimodal profiling data:

281
    ??? code
282

283
284
285
286
287
288
289
290
        ```python
        # 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)
291

292
293
294
295
        def get_dummy_mm_data(
            self,
            seq_len: int,
            mm_counts: Mapping[str, int],
296
            mm_options: Mapping[str, BaseDummyOptions],
297
298
299
300
301
302
        ) -> MultiModalDataDict:
            num_images = mm_counts.get("image", 0)

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

303
            image_overrides = mm_options.get("image")
304

305
            return {
306
307
308
309
310
311
                "image": self._get_dummy_images(
                    width=target_width,
                    height=target_height,
                    num_images=num_images,
                    overrides=image_overrides,
                )
312
313
            }
        ```
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330

    For the text, we simply expand the multimodal image token from the model config to match the desired number of images.

    ```python
    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
    ```

=== "No input placeholders: Fuyu"

    Looking at the code of HF's `FuyuForCausalLM`:

331
    ??? code
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347

        ```python
        # 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,
            )
        ```
348
349
350
351
352
353
354
355
356
357
358
359
360

    The number of placeholder feature tokens for the `i`th item in the batch is `patch_embeddings[i].shape[0]`,
    which is the same as `image_patches[i].shape[0]`, i.e. `num_total_patches`.

    Unlike LLaVA, Fuyu does not define the number of patches inside the modeling file. Where can we get more information?
    Considering that the model input comes from the output of `FuyuProcessor`, let's **look at the preprocessing files**.

    The image outputs are obtained by calling `FuyuImageProcessor.preprocess` and then
    `FuyuImageProcessor.preprocess_with_tokenizer_info` inside `FuyuProcessor`.

    In `FuyuImageProcessor.preprocess`, the images are resized and padded to the target `FuyuImageProcessor.size`,
    returning the dimensions after resizing (but before padding) as metadata.

361
    ??? code
362
363
364
365
366
367
368
369
370
371
372
373
374

        ```python
        # 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
375
376
            ]

377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
        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
            ]
        ```
396

397
    In `FuyuImageProcessor.preprocess_with_tokenizer_info`, the images are split into patches based on this metadata:
398

399
    ??? code
400
401
402
403
404
405
406
407
408
409
410

        ```python
        # 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,
411
412
        )

413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
        # 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]
        ```
434
435
436

    The number of patches is in turn defined by `FuyuImageProcessor.get_num_patches`:

437
    ??? code
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452

        ```python
        # 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
        ```
453
454
455
456
457
458
459

    These image patches correspond to placeholder tokens (`|SPEAKER|`). So, we just need to maximize the number of image patches. Since input images are first resized
    to fit within `image_processor.size`, we can maximize the number of image patches by inputting an image with size equal to `image_processor.size`.

    ```python
    def get_image_size_with_most_features(self) -> ImageSize:
        image_processor = self.get_image_processor()
460
461
462
463
        return ImageSize(
            width=image_processor.size["width"],
            height=image_processor.size["height"],
        )
464
465
466
467
468
469
470
471
472
473
474
475
    ```

    Fuyu does not expect image placeholders in the inputs to HF processor, so
    the dummy prompt text is empty regardless of the number of images.

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

    For the multimodal image profiling data, the logic is very similar to LLaVA:

476
    ??? code
477

478
479
480
481
482
        ```python
        def get_dummy_mm_data(
            self,
            seq_len: int,
            mm_counts: Mapping[str, int],
483
            mm_options: Mapping[str, BaseDummyOptions],
484
485
486
487
488
        ) -> MultiModalDataDict:
            target_width, target_height = \
                self.info.get_image_size_with_most_features()
            num_images = mm_counts.get("image", 0)

489
            image_overrides = mm_options.get("image")
490

491
            return {
492
                "image": self._get_dummy_images(
493
494
495
496
497
                    width=target_width,
                    height=target_height,
                    num_images=num_images,
                    overrides=image_overrides,
                )
498
499
            }
        ```
500
501
502
503
504
505
506

## 4. Specify processing details

Afterwards, create a subclass of [BaseMultiModalProcessor][vllm.multimodal.processing.BaseMultiModalProcessor]
to fill in the missing details about HF processing.

!!! info
507
    [Multi-Modal Data Processing](../../design/mm_processing.md)
508
509
510
511
512
513
514
515
516
517
518

### Multi-modal fields

Override [_get_mm_fields_config][vllm.multimodal.processing.BaseMultiModalProcessor._get_mm_fields_config] to
return a schema of the tensors outputted by the HF processor that are related to the input multi-modal items.

=== "Basic example: LLaVA"

    The output of `CLIPImageProcessor` is a simple tensor with shape
    `(num_images, num_channels, image_height, image_width)`:

519

520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
    ```python
    # 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)
    ```

    So, we override [_get_mm_fields_config][vllm.multimodal.processing.BaseMultiModalProcessor._get_mm_fields_config] as follows:

    ```python
    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"),
        )
    ```

    !!! note
545
        Our [actual code](../../../vllm/model_executor/models/llava.py) additionally supports
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
        pre-computed image embeddings, which can be passed to be model via the `image_embeds` argument.

=== "With postprocessing: Fuyu"

    The `image_patches` output of `FuyuImageProcessor.preprocess_with_tokenizer_info` concatenates
    the patches from each image belonging to an item in the batch:

    ```python
    # 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)
    ```

    The shape of `image_patches` outputted by `FuyuImageProcessor` is therefore
    `(1, num_images, num_patches, patch_width * patch_height * num_channels)`.

567
568
569
570
    In order to support the use of
    [MultiModalFieldConfig.batched][vllm.multimodal.inputs.MultiModalFieldConfig.batched]
    like in LLaVA, we remove the extra batch dimension by overriding
    [BaseMultiModalProcessor._call_hf_processor][vllm.multimodal.processing.BaseMultiModalProcessor._call_hf_processor]:
571

572
    ??? code
573

574
575
576
577
578
579
        ```python
        def _call_hf_processor(
            self,
            prompt: str,
            mm_data: Mapping[str, object],
            mm_kwargs: Mapping[str, object],
580
            tok_kwargs: Mapping[str, object],
581
582
583
584
585
        ) -> BatchFeature:
            processed_outputs = super()._call_hf_processor(
                prompt=prompt,
                mm_data=mm_data,
                mm_kwargs=mm_kwargs,
586
                tok_kwargs=tok_kwargs,
587
            )
588

589
590
591
592
            image_patches = processed_outputs.get("image_patches")
            if image_patches is not None:
                images = mm_data["images"]
                assert isinstance(images, list)
593

594
595
596
597
598
599
                # 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))
600

601
602
603
604
                processed_outputs["image_patches"] = image_patches[0]

            return processed_outputs
        ```
605
606

    !!! note
607
        Our [actual code](../../../vllm/model_executor/models/fuyu.py) has special handling
608
609
        for text-only inputs to prevent unnecessary warnings from HF processor.

610
611
612
613
614
    !!! note
        The `_call_hf_processor` method specifies both `mm_kwargs` and `tok_kwargs` for
        processing. `mm_kwargs` is used to both initialize and call the huggingface
        processor, whereas `tok_kwargs` is only used to call the huggingface processor.

615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
    This lets us override [_get_mm_fields_config][vllm.multimodal.processing.BaseMultiModalProcessor._get_mm_fields_config] as follows:

    ```python
    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 updates

Override [_get_prompt_updates][vllm.multimodal.processing.BaseMultiModalProcessor._get_prompt_updates] to
return a list of [PromptUpdate][vllm.multimodal.processing.PromptUpdate] instances.

Each [PromptUpdate][vllm.multimodal.processing.PromptUpdate] instance specifies an update operation
(e.g.: insertion, replacement) performed by the HF processor.

=== "Basic example: LLaVA"

    Looking at HF's `LlavaProcessor`:

    ```python
    # 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)
    ```

    It simply repeats each input `image_token` a number of times equal to the number of placeholder feature tokens (`num_image_tokens`).
    Based on this, we override [_get_prompt_updates][vllm.multimodal.processing.BaseMultiModalProcessor._get_prompt_updates] as follows:

649
    ??? code
650

651
652
653
654
655
        ```python
        def _get_prompt_updates(
            self,
            mm_items: MultiModalDataItems,
            hf_processor_mm_kwargs: Mapping[str, object],
656
            out_mm_kwargs: MultiModalKwargsItems,
657
658
659
660
661
662
663
664
665
666
667
668
        ) -> 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,
                )
669

670
                return [image_token_id] * num_image_tokens
671

672
673
674
675
676
677
678
679
            return [
                PromptReplacement(
                    modality="image",
                    target=[image_token_id],
                    replacement=get_replacement,
                ),
            ]
        ```
680
681
682
683
684
685
686
687
688
689
690
691
692
693

=== "Handling additional tokens: Fuyu"

    Recall the layout of feature tokens from Step 2:

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

    We define a helper function to return `ncols` and `nrows` directly:

694
    ??? code
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720

        ```python
        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
        ```
721
722
723

    Based on this, we can initially define our replacement tokens as:

724
    ??? code
725

726
727
728
729
        ```python
        def get_replacement(item_idx: int):
            images = mm_items.get_items("image", ImageProcessorItems)
            image_size = images.get_image_size(item_idx)
730

731
732
733
734
735
736
737
738
739
            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
        ```
740
741

    However, this is not entirely correct. After `FuyuImageProcessor.preprocess_with_tokenizer_info` is called,
742
    a BOS token (`<s>`) is also added to the prompt:
743

744
    ??? code
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766

        ```python
        # 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,
        )
        ```
767
768
769
770

    To assign the vision embeddings to only the image tokens, instead of a string
    you can return an instance of [PromptUpdateDetails][vllm.multimodal.processing.PromptUpdateDetails]:

771
    ??? code
772

773
        ```python
774
        hf_config = self.info.get_hf_config()
775
        bos_token_id = hf_config.bos_token_id  # `<s>`
776
777
778
779
780
781
782
783
784
785
        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,
            )
786
            image_tokens = ([_IMAGE_TOKEN_ID] * ncols + [_NEWLINE_TOKEN_ID]) * nrows
787
788
789
790
791

            return PromptUpdateDetails.select_token_id(
                image_tokens + [bos_token_id],
                embed_token_id=_IMAGE_TOKEN_ID,
            )
792
        ```
793

794
795
796
    Finally, noticing that the HF processor removes the `|ENDOFTEXT|` token from the tokenized prompt,
    we can search for it to conduct the replacement at the start of the string:

797
    ??? code
798
799
800
801
802
803

        ```python
        def _get_prompt_updates(
            self,
            mm_items: MultiModalDataItems,
            hf_processor_mm_kwargs: Mapping[str, object],
804
            out_mm_kwargs: MultiModalKwargsItems,
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
        ) -> 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,
                )
822
                image_tokens = ([_IMAGE_TOKEN_ID] * ncols + [_NEWLINE_TOKEN_ID]) * nrows
823
824
825
826
827
828
829
830
831
832
833
834
835
836

                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,
                )
            ]
        ```
837
838
839
840

## 5. Register processor-related classes

After you have defined [BaseProcessingInfo][vllm.multimodal.processing.BaseProcessingInfo] (Step 2),
841
[BaseDummyInputsBuilder][vllm.multimodal.processing.BaseDummyInputsBuilder] (Step 3),
842
and [BaseMultiModalProcessor][vllm.multimodal.processing.BaseMultiModalProcessor] (Step 4),
843
decorate the model class with [MULTIMODAL_REGISTRY.register_processor][vllm.multimodal.registry.MultiModalRegistry.register_processor]
844
845
846
847
848
849
to register them to the multi-modal registry:

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

850
851
852
853
854
+ @MULTIMODAL_REGISTRY.register_processor(
+     YourMultiModalProcessor,
+     info=YourProcessingInfo,
+     dummy_inputs=YourDummyInputsBuilder,
+ )
855
856
857
858
859
860
861
862
863
864
865
  class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
```

## Notes

### Inserting feature tokens without replacement

Some HF processors directly insert feature tokens without replacing anything in the original prompt. In that case, you can use [PromptInsertion][vllm.multimodal.processing.PromptInsertion] instead of [PromptReplacement][vllm.multimodal.processing.PromptReplacement] inside [_get_prompt_updates][vllm.multimodal.processing.BaseMultiModalProcessor._get_prompt_updates].

Examples:

866
867
- BLIP-2 (insert at start of prompt): [vllm/model_executor/models/blip2.py](../../../vllm/model_executor/models/blip2.py)
- Molmo (insert after `<|endoftext|>` token): [vllm/model_executor/models/molmo.py](../../../vllm/model_executor/models/molmo.py)
868
869
870

### Handling prompt updates unrelated to multi-modal data

871
[_get_prompt_updates][vllm.multimodal.processing.BaseMultiModalProcessor._get_prompt_updates] assumes that each application of prompt update corresponds to one multi-modal item. If the HF processor performs additional processing regardless of how many multi-modal items there are, you should override [_apply_hf_processor_tokens_only][vllm.multimodal.processing.BaseMultiModalProcessor._apply_hf_processor_tokens_only] so that the processed token inputs are consistent with the result of applying the HF processor on text inputs. This is because token inputs bypass the HF processor according to [our design](../../design/mm_processing.md).
872
873
874

Examples:

875
876
877
- Chameleon (appends `sep_token`): [vllm/model_executor/models/chameleon.py](../../../vllm/model_executor/models/chameleon.py)
- Fuyu (appends `boa_token`): [vllm/model_executor/models/fuyu.py](../../../vllm/model_executor/models/fuyu.py)
- Molmo (applies chat template which is not defined elsewhere): [vllm/model_executor/models/molmo.py](../../../vllm/model_executor/models/molmo.py)
878
879
880

### Custom HF processor

881
Some models don't define an HF processor class on HF Hub. In that case, you can define a custom HF processor that has the same call signature as HF processors and pass it to [_call_hf_processor][vllm.multimodal.processing.BaseMultiModalProcessor._call_hf_processor].
882
883
884

Examples:

885
886
- DeepSeek-VL2: [vllm/model_executor/models/deepseek_vl2.py](../../../vllm/model_executor/models/deepseek_vl2.py)
- InternVL: [vllm/model_executor/models/internvl.py](../../../vllm/model_executor/models/internvl.py)
zhuwenwen's avatar
zhuwenwen committed
887
- Qwen-VL: [vllm/model_executor/models/qwen_vl.py](../../../vllm/model_executor/models/qwen_vl.py)