[Multi Modal] Configurable MM Profiling (#25631)

Signed-off-by: wwl2755 <wangwenlong2755@gmail.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

vllm/model_executor/models/blip2.py

@@ -10,6 +10,7 @@ from transformers import (BatchFeature, Blip2Config, Blip2QFormerConfig,
                           apply_chunking_to_forward)
 
 from vllm.config import CacheConfig, VllmConfig
+from vllm.config.multimodal import BaseDummyOptions
 from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.multimodal import MULTIMODAL_REGISTRY
@@ -435,6 +436,7 @@ class Blip2DummyInputsBuilder(BaseDummyInputsBuilder[Blip2ProcessingInfo]):
         self,
         seq_len: int,
         mm_counts: Mapping[str, int],
+        mm_options: Optional[Mapping[str, BaseDummyOptions]] = None,
     ) -> MultiModalDataDict:
         hf_config = self.info.get_hf_config()
         vision_config = hf_config.vision_config
@@ -442,11 +444,14 @@ class Blip2DummyInputsBuilder(BaseDummyInputsBuilder[Blip2ProcessingInfo]):
         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
+
         return {
             "image":
             self._get_dummy_images(width=max_image_size,
                                    height=max_image_size,
-                                   num_images=num_images)
+                                   num_images=num_images,
+                                   overrides=image_overrides)
         }