Simplify code

vllm/inputs/registry.py

@@ -185,7 +185,7 @@ def dummy_data_for_profiling(
         mm_counts = mm_registry.get_mm_limits_per_prompt(model_config)
 
         mm_processor_kwargs = get_allowed_kwarg_only_overrides(
-            callable=dummy_factory, overrides=model_config.mm_processor_kwargs)
+            dummy_factory, overrides=model_config.mm_processor_kwargs)
 
         seq_data, mm_data = dummy_factory(InputContext(model_config), seq_len,
                                           _MultiModalCounts(mm_counts),
@@ -255,7 +255,7 @@ def process_input(self, model_config: "ModelConfig",
             .get(model_cls, self._default_input_processor)
 
         mm_processor_kwargs = get_allowed_kwarg_only_overrides(
-            callable=processor, overrides=model_config.mm_processor_kwargs)
+            processor, overrides=model_config.mm_processor_kwargs)
 
         return processor(InputContext(model_config), inputs,
                          **mm_processor_kwargs)

vllm/multimodal/base.py

@@ -260,9 +260,9 @@ def map_input(self, model_config: ModelConfig,
         # Only get processor kwargs at mapping time if we are not using the
         # input mapper; no overrides are used on the default here because they
         # should be passed to the huggingface resource at initialization time.
-        if mapper != self._default_input_mapper:
+        if mapper is not None and mapper != self._default_input_mapper:
             mm_processor_kwargs = get_allowed_kwarg_only_overrides(
-                callable=mapper, overrides=model_config.mm_processor_kwargs)
+                mapper, overrides=model_config.mm_processor_kwargs)
         else:
             mm_processor_kwargs = {}
 
@@ -344,8 +344,7 @@ def get_max_multimodal_tokens(self, model_config: ModelConfig) -> int:
 
         if callable(max_mm_tokens):
             mm_processor_kwargs = get_allowed_kwarg_only_overrides(
-                callable=max_mm_tokens,
-                overrides=model_config.mm_processor_kwargs)
+                max_mm_tokens, overrides=model_config.mm_processor_kwargs)
             max_mm_tokens = max_mm_tokens(InputContext(model_config),
                                           **mm_processor_kwargs)
 

vllm/utils.py

@@ -1239,7 +1239,7 @@ async def _run_task_with_lock(task: Callable, lock: asyncio.Lock, *args,
 
 
 def get_allowed_kwarg_only_overrides(
-    callable: Optional[Callable],
+    callable: Callable[..., object],
     overrides: Optional[Dict[str, Any]],
 ) -> Dict[str, Any]:
     """
@@ -1259,7 +1259,7 @@ def get_allowed_kwarg_only_overrides(
         to overwrite one or more keyword only arguments when invoking the
         callable.
     """
-    if not overrides or not callable:
+    if not overrides:
         return {}
 
     allowed_override_names = [
@@ -1276,7 +1276,7 @@ def get_allowed_kwarg_only_overrides(
     }
 
     # If anything is dropped, log a warning
-    dropped_keys = set(overrides) - set(filtered_overrides)
+    dropped_keys = overrides.keys() - filtered_overrides.keys()
     if dropped_keys:
         logger.warning(
             "The following intended overrides are not keyword-only args "