[Refactor] Simplify dummy data generation (#35025)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/model_executor/models/funasr.py

@@ -746,23 +746,22 @@ class FunASRDummyInputsBuilder(BaseDummyInputsBuilder[FunASRProcessingInfo]):
         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 = feature_extractor.chunk_length * 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(
-                length=audio_len, num_audios=num_audios, overrides=audio_overrides
-            )
+                length=audio_len,
+                num_audios=num_audios,
+                overrides=audio_overrides,
+            ),
         }
 
 

vllm/model_executor/models/funaudiochat.py

@@ -610,12 +610,9 @@ class FunAudioChatDummyInputsBuilder(
         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 = int(feature_extractor.sampling_rate)
 
         # Dummy inputs are used for profiling; construct the worst-case audio
@@ -632,7 +629,7 @@ class FunAudioChatDummyInputsBuilder(
         )
         num_audios = int(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(
                 length=audio_len,

vllm/model_executor/models/fuyu.py

@@ -142,13 +142,12 @@ class FuyuDummyInputsBuilder(BaseDummyInputsBuilder[FuyuProcessingInfo]):
         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:
         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(

vllm/model_executor/models/gemma3_mm.py

@@ -241,14 +241,13 @@ class Gemma3DummyInputsBuilder(BaseDummyInputsBuilder[Gemma3ProcessingInfo]):
         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/gemma3n_mm.py

@@ -175,8 +175,7 @@ class Gemma3nDummyInputsBuilder(BaseDummyInputsBuilder[Gemma3nProcessingInfo]):
         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)
         num_audios = mm_counts.get("audio", 0)
@@ -189,8 +188,8 @@ class Gemma3nDummyInputsBuilder(BaseDummyInputsBuilder[Gemma3nProcessingInfo]):
         img_width = image_processor.size.get("width", 224)
         img_height = image_processor.size.get("height", 224)
 
-        image_overrides = mm_options.get("image") if mm_options else None
-        audio_overrides = mm_options.get("audio") if mm_options else None
+        image_overrides = mm_options.get("image")
+        audio_overrides = mm_options.get("audio")
 
         return {
             "image": self._get_dummy_images(
@@ -200,7 +199,9 @@ class Gemma3nDummyInputsBuilder(BaseDummyInputsBuilder[Gemma3nProcessingInfo]):
                 overrides=image_overrides,
             ),
             "audio": self._get_dummy_audios(
-                length=audio_len, num_audios=num_audios, overrides=audio_overrides
+                length=audio_len,
+                num_audios=num_audios,
+                overrides=audio_overrides,
             ),
         }
 

vllm/model_executor/models/glm4_1v.py

@@ -1163,8 +1163,7 @@ class Glm4vDummyInputsBuilder(BaseDummyInputsBuilder[Glm4vProcessingInfo]):
         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)
         num_videos = mm_counts.get("video", 0)
@@ -1174,8 +1173,8 @@ class Glm4vDummyInputsBuilder(BaseDummyInputsBuilder[Glm4vProcessingInfo]):
             seq_len, mm_counts
         )
 
-        image_overrides = mm_options.get("image") if mm_options else None
-        video_overrides = mm_options.get("video") if mm_options else None
+        image_overrides = mm_options.get("image")
+        video_overrides = mm_options.get("video")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/glm4v.py

@@ -492,8 +492,7 @@ class GLM4VDummyInputsBuilder(BaseDummyInputsBuilder[GLM4VProcessingInfo]):
         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
@@ -501,7 +500,7 @@ class GLM4VDummyInputsBuilder(BaseDummyInputsBuilder[GLM4VProcessingInfo]):
         target_width = target_height = 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/glmasr.py

@@ -726,15 +726,12 @@ class GlmAsrDummyInputsBuilder(BaseDummyInputsBuilder[GlmAsrProcessingInfo]):
         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
         num_audios = mm_counts.get("audio", 0)
-        audio_overrides = mm_options.get("audio") if mm_options else None
+        audio_overrides = mm_options.get("audio")
 
         max_audio_len = getattr(
             self.info.get_hf_processor(), "max_audio_len", DEFAULT_MAX_AUDIO_LEN_S
@@ -743,7 +740,9 @@ class GlmAsrDummyInputsBuilder(BaseDummyInputsBuilder[GlmAsrProcessingInfo]):
 
         return {
             "audio": self._get_dummy_audios(
-                length=audio_len, num_audios=num_audios, overrides=audio_overrides
+                length=audio_len,
+                num_audios=num_audios,
+                overrides=audio_overrides,
             )
         }
 

vllm/model_executor/models/granite_speech.py

@@ -216,11 +216,10 @@ class GraniteSpeechDummyInputsBuilder(
         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_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/hunyuan_vision.py

@@ -713,8 +713,7 @@ class HunYuanVLDummyInputsBuilder(BaseDummyInputsBuilder[HunYuanVLProcessingInfo
         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", 1)
 

vllm/model_executor/models/hyperclovax_vision.py

@@ -165,8 +165,7 @@ class HCXVisionDummyInputsBuilder(BaseDummyInputsBuilder[HCXVisionProcessingInfo
         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)
         num_videos = mm_counts.get("video", 0)
@@ -174,8 +173,8 @@ class HCXVisionDummyInputsBuilder(BaseDummyInputsBuilder[HCXVisionProcessingInfo
         target_width, target_height = self.info.get_image_size_with_most_features()
         target_num_frames = 32
 
-        image_overrides = mm_options.get("image") if mm_options else None
-        video_overrides = mm_options.get("video") if mm_options else None
+        image_overrides = mm_options.get("image")
+        video_overrides = mm_options.get("video")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/idefics3.py

@@ -277,15 +277,14 @@ class Idefics3DummyInputsBuilder(BaseDummyInputsBuilder[Idefics3ProcessingInfo])
         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_processor = self.info.get_hf_processor(**(mm_processor_kwargs or {}))
+        hf_processor = self.info.get_hf_processor()
         image_processor: Idefics3ImageProcessor = hf_processor.image_processor
         longest_edge = image_processor.max_image_size["longest_edge"]
 
-        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/interns1.py

@@ -297,8 +297,7 @@ class InternS1DummyInputsBuilder(BaseDummyInputsBuilder[InternS1ProcessingInfo])
         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:
         target_width, target_height = self.info.get_image_size_with_most_features()
         target_num_frames = self.info.get_num_frames_with_most_features(
@@ -310,8 +309,8 @@ class InternS1DummyInputsBuilder(BaseDummyInputsBuilder[InternS1ProcessingInfo])
         config = self.info.get_hf_config()
         image_size_h, image_size_w = config.vision_config.image_size
 
-        image_overrides = mm_options.get("image") if mm_options else None
-        video_overrides = mm_options.get("video") if mm_options else None
+        image_overrides = mm_options.get("image")
+        video_overrides = mm_options.get("video")
 
         return {
             "image": self._get_dummy_images(

vllm/model_executor/models/internvl.py

@@ -762,13 +762,12 @@ class BaseInternVLDummyInputsBuilder(BaseDummyInputsBuilder[_I]):
         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:
         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(
@@ -935,12 +934,9 @@ class InternVLDummyInputsBuilder(
         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:
-        dummy_image = super().get_dummy_mm_data(
-            seq_len=seq_len, mm_counts=mm_counts, mm_options=mm_options
-        )
+        dummy_image = super().get_dummy_mm_data(seq_len, mm_counts, mm_options)
         if self.info.supports_video:
             config = self.info.get_hf_config()
             image_size: int = config.vision_config.image_size
@@ -948,7 +944,7 @@ class InternVLDummyInputsBuilder(
                 seq_len, mm_counts
             )
             num_videos = mm_counts.get("video", 0)
-            video_overrides = mm_options.get("video") if mm_options else None
+            video_overrides = mm_options.get("video")
             dummy_video = {
                 "video": self._get_dummy_videos(
                     width=image_size,

vllm/model_executor/models/isaac.py

@@ -18,6 +18,7 @@ from typing_extensions import TypedDict, Unpack
 
 from vllm.config import VllmConfig
 from vllm.config.model import ModelConfig
+from vllm.config.multimodal import BaseDummyOptions
 from vllm.distributed import parallel_state
 from vllm.distributed import utils as dist_utils
 from vllm.model_executor.layers.attention import MMEncoderAttention
@@ -849,13 +850,12 @@ class IsaacDummyInputsBuilder(BaseDummyInputsBuilder[IsaacProcessingInfo]):
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
-        mm_options: Mapping[str] | 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/kanana_v.py

@@ -444,8 +444,7 @@ class KananaVDummyInputsBuilder(BaseDummyInputsBuilder[KananaVProcessingInfo]):
         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)
         return {

vllm/model_executor/models/keye.py

@@ -1170,8 +1170,7 @@ class KeyeBaseDummyInputsBuilder(BaseDummyInputsBuilder[_I]):
         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)
         num_videos = mm_counts.get("video", 0)
@@ -1179,8 +1178,8 @@ class KeyeBaseDummyInputsBuilder(BaseDummyInputsBuilder[_I]):
         target_width, target_height = self.info.get_image_size_with_most_features()
         target_num_frames = self.info.get_num_frames_with_most_features(seq_len)
 
-        image_overrides = mm_options.get("image") if mm_options else None
-        video_overrides = mm_options.get("video") if mm_options else None
+        image_overrides = mm_options.get("image")
+        video_overrides = mm_options.get("video")
 
         mm_data = {
             "image": self._get_dummy_images(

vllm/model_executor/models/kimi_k25.py

@@ -240,8 +240,7 @@ class KimiK25DummyInputsBuilder(BaseDummyInputsBuilder[KimiK25ProcessingInfo]):
         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:
         # TODO: Support mm_options for vision_chunk to allow user configuration
         dummy_items = self.get_dummy_mm_items()

vllm/model_executor/models/kimi_vl.py

@@ -215,12 +215,11 @@ class KimiVLDummyInputsBuilder(BaseDummyInputsBuilder[KimiVLProcessingInfo]):
         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_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/lfm2_vl.py

@@ -343,14 +343,13 @@ class Lfm2VLDummyInputsBuilder(BaseDummyInputsBuilder[Lfm2VLProcessingInfo]):
         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(