[1/N] Initial prototype for multi-modal processor (#10044)

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

vllm/multimodal/audio.py

@@ -1,5 +1,7 @@
 from vllm.inputs.registry import InputContext
-from vllm.multimodal.base import MultiModalKwargs, MultiModalPlugin
+
+from .base import MultiModalPlugin
+from .inputs import AudioItem, MultiModalData, MultiModalKwargs
 
 
 class AudioPlugin(MultiModalPlugin):
@@ -8,8 +10,12 @@ class AudioPlugin(MultiModalPlugin):
     def get_data_key(self) -> str:
         return "audio"
 
-    def _default_input_mapper(self, ctx: InputContext, data: object,
-                              **mm_processor_kwargs) -> MultiModalKwargs:
+    def _default_input_mapper(
+        self,
+        ctx: InputContext,
+        data: MultiModalData[AudioItem],
+        **mm_processor_kwargs,
+    ) -> MultiModalKwargs:
         raise NotImplementedError("There is no default audio input mapper")
 
     def _default_max_multimodal_tokens(self, ctx: InputContext) -> int: