[VLM] Reorganize profiling/processing-related code (#11812)

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

vllm/inputs/preprocess.py

@@ -7,7 +7,7 @@ from vllm.config import ModelConfig
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
-from vllm.multimodal.processing import MultiModalDataDict, MultiModalInputsV2
+from vllm.multimodal.inputs import MultiModalDataDict, MultiModalInputsV2
 from vllm.prompt_adapter.request import PromptAdapterRequest
 from vllm.transformers_utils.tokenizer_group import BaseTokenizerGroup
 from vllm.utils import print_info_once, print_warning_once

vllm/inputs/registry.py

@@ -323,6 +323,7 @@ class InputRegistry:
         # Avoid circular import
         from vllm.model_executor.model_loader import get_model_architecture
         from vllm.multimodal import MultiModalKwargs
+        from vllm.multimodal.profiling import MultiModalProfiler
         from vllm.multimodal.utils import cached_get_tokenizer
 
         if mm_registry.has_processor(model_config):
@@ -331,7 +332,8 @@ class InputRegistry:
                 trust_remote_code=model_config.trust_remote_code,
             )
             processor = mm_registry.create_processor(model_config, tokenizer)
-            dummy_data = processor.get_dummy_data(seq_len)
+            profiler = MultiModalProfiler(processor)
+            dummy_data = profiler.get_dummy_data(seq_len)
         else:
             model_cls, _ = get_model_architecture(model_config)
             if is_encoder_data: