[Refactor] Simplify dummy data generation

docs/contributing/model/multimodal.md

@@ -293,21 +293,22 @@ Assuming that the memory usage increases with the number of tokens, the dummy in
             self,
             seq_len: int,
             mm_counts: Mapping[str, int],
-            mm_options: Mapping[str, BaseDummyOptions] | None = None,
+            mm_options: Mapping[str, BaseDummyOptions],
         ) -> MultiModalDataDict:
             num_images = mm_counts.get("image", 0)
 
             target_width, target_height = \
                 self.info.get_image_size_with_most_features()
 
-            image_overrides = mm_options.get("image") if mm_options else None
+            image_overrides = mm_options.get("image")
 
             return {
-                "image":
-                self._get_dummy_images(width=target_width,
-                                    height=target_height,
-                                    num_images=num_images,
-                                    overrides=image_overrides)
+                "image": self._get_dummy_images(
+                    width=target_width,
+                    height=target_height,
+                    num_images=num_images,
+                    overrides=image_overrides,
+                )
             }
         ```
 
@@ -479,17 +480,16 @@ Assuming that the memory usage increases with the number of tokens, the dummy in
             self,
             seq_len: int,
             mm_counts: Mapping[str, int],
-            mm_options: Optional[Mapping[str, BaseDummyOptions]] = None,
+            mm_options: Mapping[str, BaseDummyOptions],
         ) -> MultiModalDataDict:
             target_width, target_height = \
                 self.info.get_image_size_with_most_features()
             num_images = mm_counts.get("image", 0)
 
-            image_overrides = mm_options.get("image") if mm_options else None
+            image_overrides = mm_options.get("image")
 
             return {
-                "image":
-                self._get_dummy_images(
+                "image": self._get_dummy_images(
                     width=target_width,
                     height=target_height,
                     num_images=num_images,

tests/models/multimodal/processing/test_audioflamingo3.py

@@ -116,7 +116,7 @@ def test_dummy_data_generation(mock_ctx):
     builder = AudioFlamingo3DummyInputsBuilder(info)
 
     mm_counts = {"audio": 2}
-    dummy_data = builder.get_dummy_mm_data(100, mm_counts, None)
+    dummy_data = builder.get_dummy_mm_data(100, mm_counts, {})
 
     assert "audio" in dummy_data
     assert len(dummy_data["audio"]) == 2

vllm/config/multimodal.py

@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from collections.abc import Mapping
-from typing import Any, Literal, TypeAlias
+from typing import Any, Literal, TypeAlias, TypedDict, final
 
 from pydantic import ConfigDict, Field, field_validator, model_validator
 from pydantic.dataclasses import dataclass
@@ -43,11 +43,29 @@ class AudioDummyOptions(BaseDummyOptions):
     length: int | None = Field(None, gt=0)
 
 
+@final
+class MultiModalDummyOptionsBuiltins(TypedDict, total=False):
+    """Type annotations for modality types predefined by vLLM."""
+
+    image: ImageDummyOptions
+    """Options for dummy images."""
+
+    video: VideoDummyOptions
+    """Options for dummy videos."""
+
+    audio: AudioDummyOptions
+    """Options for dummy audios."""
+
+
 MMEncoderTPMode = Literal["weights", "data"]
 MMCacheType = Literal["shm", "lru"]
-DummyOptions: TypeAlias = (
-    BaseDummyOptions | VideoDummyOptions | ImageDummyOptions | AudioDummyOptions
-)
+MMDummyOptions: TypeAlias = dict[str, BaseDummyOptions]
+"""
+A dictionary containing an entry for each modality type of dummy data.
+
+The built-in modalities are defined by
+[`MultiModalDummyOptionsBuiltins`][vllm.config.multimodal.MultiModalDummyOptionsBuiltins].
+"""
 
 
 @config
@@ -57,7 +75,7 @@ class MultiModalConfig:
     language_model_only: bool = False
     """If True, disables all multimodal inputs by setting all modality limits to 0.
     Equivalent to setting `--limit-mm-per-prompt` to 0 for every modality."""
-    limit_per_prompt: dict[str, DummyOptions] = Field(default_factory=dict)
+    limit_per_prompt: MMDummyOptions = Field(default_factory=dict)
     """The maximum number of input items and options allowed per
     prompt for each modality.
 
@@ -158,22 +176,27 @@ class MultiModalConfig:
     @field_validator("limit_per_prompt", mode="before")
     @classmethod
     def _validate_limit_per_prompt(
-        cls, value: dict[str, int | dict[str, int]]
-    ) -> dict[str, DummyOptions]:
+        cls,
+        value: dict[str, int | dict[str, int]],
+    ) -> MMDummyOptions:
+        out: MMDummyOptions = {}
+
         for k, v in value.items():
             # Handle legacy format where only count is specified
             if isinstance(v, int):
                 v = {"count": v}
+
             # Convert to the appropriate DummyOptions subclass
             if k == "video":
-                value[k] = VideoDummyOptions(**v)
+                out[k] = VideoDummyOptions(**v)
             elif k == "image":
-                value[k] = ImageDummyOptions(**v)
+                out[k] = ImageDummyOptions(**v)
             elif k == "audio":
-                value[k] = AudioDummyOptions(**v)
+                out[k] = AudioDummyOptions(**v)
             else:
-                value[k] = BaseDummyOptions(**v)
-        return value
+                out[k] = BaseDummyOptions(**v)
+
+        return out
 
     @field_validator("mm_encoder_attn_backend", mode="before")
     @classmethod
@@ -240,15 +263,8 @@ def get_limit_per_prompt(self, modality: str) -> int:
         if limit_data is None:
             # Unspecified modality is set to 999 by default
             return 999
-        return limit_data.count
 
-    def get_dummy_options(self, modality: str) -> BaseDummyOptions | None:
-        """
-        Get the configurable dummy data options for a modality.
-        Returns None if no options are configured for this modality.
-        """
-        # All values are now DummyOptions after normalization
-        return self.limit_per_prompt.get(modality)
+        return limit_data.count
 
     def merge_mm_processor_kwargs(
         self,

vllm/model_executor/models/aria.py

@@ -444,15 +444,14 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         vision_config = self.info.get_vision_config()
 
         max_image_size = vision_config.image_size
         num_images = mm_counts.get("image", 0)
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/audioflamingo3.py

@@ -252,16 +252,13 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
-        feature_extractor = self.info.get_feature_extractor(
-            **(mm_processor_kwargs or {})
-        )
+        feature_extractor = self.info.get_feature_extractor()
         sampling_rate = feature_extractor.sampling_rate
         audio_len = MAX_AUDIO_LEN * sampling_rate
         num_audios = mm_counts.get("audio", 0)
-        audio_overrides = mm_options.get("audio") if mm_options else None
+        audio_overrides = mm_options.get("audio")
 
         return {
             "audio": self._get_dummy_audios(

vllm/model_executor/models/aya_vision.py

@@ -191,13 +191,12 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
         image_size = self.info.get_image_size_with_most_features()
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/bagel.py

@@ -249,16 +249,15 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
         hf_config = self.info.get_hf_config()
         vit_config = hf_config.vit_config
 
         # Use the configured image size
         image_size = vit_config.image_size
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/bee.py

@@ -90,14 +90,13 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
 
         target_width, target_height = self.info.get_image_size_with_most_features()
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/blip2.py

@@ -445,16 +445,15 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         hf_config = self.info.get_hf_config()
         vision_config = hf_config.vision_config
 
         max_image_size = vision_config.image_size
         num_images = mm_counts.get("image", 0)
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/chameleon.py

@@ -116,15 +116,14 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         config = self.info.get_hf_config()
 
         width = height = config.vq_config.resolution
         num_images = mm_counts.get("image", 0)
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/clip.py

@@ -174,14 +174,13 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
 
         target_width, target_height = self.info.get_image_size_with_most_features()
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/cohere2_vision.py

@@ -197,13 +197,12 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
         image_size = self.info.get_image_size_with_most_features()
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/deepseek_ocr.py

@@ -255,8 +255,7 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
 

vllm/model_executor/models/deepseek_ocr2.py

@@ -137,8 +137,7 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
 

vllm/model_executor/models/deepseek_vl2.py

@@ -214,14 +214,13 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
 
         max_image_size = self.info.get_image_size_with_most_features()
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/dots_ocr.py

@@ -106,17 +106,13 @@ def get_dummy_mm_data(
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str, BaseDummyOptions] | None = None,
-        mm_processor_kwargs: Mapping[str, object] | None = None,
+        mm_options: Mapping[str, BaseDummyOptions],
     ) -> MultiModalDataDict:
         num_images = mm_counts.get("image", 0)
 
-        mm_processor_kwargs = mm_processor_kwargs or {}
-        target_width, target_height = self.info.get_image_size_with_most_features(  # noqa: E501
-            mm_processor_kwargs.get("max_pixels", None)
-        )
+        target_width, target_height = self.info.get_image_size_with_most_features()
 
-        image_overrides = mm_options.get("image") if mm_options else None
+        image_overrides = mm_options.get("image")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/eagle2_5_vl.py

@@ -149,7 +149,7 @@ class Eagle2_5_VLProcessingInfo(BaseInternVLProcessingInfo):
     """Processing info for Eagle2.5-VL model."""
 
     def get_hf_processor(self, **kwargs) -> Eagle2_5_VLProcessor:
-        return self.ctx.init_processor(
+        return self.ctx.get_hf_processor(
             Eagle2_5_VLProcessor,
             config=self.ctx.get_hf_config(),
             tokenizer=self.get_tokenizer(),