[Refactor] Decouple TimingContext from InputProcessingContext (#35083)

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

vllm/model_executor/models/deepseek_vl2.py

@@ -30,15 +30,16 @@ from vllm.multimodal.parse import (
     ImageProcessorItems,
     ImageSize,
     MultiModalDataItems,
-    MultiModalUUIDItems,
 )
 from vllm.multimodal.processing import BaseDummyInputsBuilder
 from vllm.multimodal.processing.processor import (
     BaseMultiModalProcessor,
     BaseProcessingInfo,
     MultiModalProcessingInfo,
+    ProcessorInputs,
     PromptReplacement,
     PromptUpdate,
+    TimingContext,
 )
 from vllm.sequence import IntermediateTensors
 from vllm.tokenizers import cached_tokenizer_from_config
@@ -310,32 +311,17 @@ class DeepseekVL2MultiModalProcessor(
 
     def _cached_apply_hf_processor(
         self,
-        prompt: str | list[int],
-        mm_data_items: MultiModalDataItems,
-        mm_uuid_items: MultiModalUUIDItems | None,
-        hf_processor_mm_kwargs: Mapping[str, object],
-        tokenization_kwargs: Mapping[str, object],
+        inputs: ProcessorInputs,
+        timing_ctx: TimingContext,
     ) -> tuple[list[int], MultiModalProcessingInfo, bool]:
         # The processor logic is different for len(images) <= 2 vs > 2
         # Since the processing cache assumes that the processor output is
         # invariant of how many images are passed per prompt, we only
         # perform caching for the most common case
-        if mm_data_items.get_count("image", strict=False) > 2:
-            return self._apply_hf_processor(
-                prompt=prompt,
-                mm_data_items=mm_data_items,
-                mm_uuid_items=mm_uuid_items,
-                hf_processor_mm_kwargs=hf_processor_mm_kwargs,
-                tokenization_kwargs=tokenization_kwargs,
-            )
+        if inputs.mm_data_items.get_count("image", strict=False) > 2:
+            return self._apply_hf_processor(inputs, timing_ctx)
 
-        return super()._cached_apply_hf_processor(
-            prompt=prompt,
-            mm_data_items=mm_data_items,
-            mm_uuid_items=mm_uuid_items,
-            hf_processor_mm_kwargs=hf_processor_mm_kwargs,
-            tokenization_kwargs=tokenization_kwargs,
-        )
+        return super()._cached_apply_hf_processor(inputs, timing_ctx)
 
 
 @MULTIMODAL_REGISTRY.register_processor(