[V1] Avoid sending text prompt to core engine (vllm-project#11963)

Signed-off-by: Roger Wang <[email protected]>

Author: ywang96

vllm/v1/engine/__init__.py

@@ -19,8 +19,8 @@ class EngineCoreRequest:
     # due to circular imports and typing we have in data.py
 
     request_id: str
-    #NOTE(Nick): I don't think we need to pass prompt here since it should
-    # always be tokenized?
+    # NOTE(ywang96): original text prompt is needed when a request is added to
+    # Detokenizer, but set to None when it is added to EngineCoreClient.
     prompt: Optional[str]
     prompt_token_ids: List[int]
     mm_inputs: Optional[List[Optional["MultiModalKwargs"]]]

vllm/v1/engine/core_client.py

@@ -219,6 +219,9 @@ def _send_input(self, request_type: EngineCoreRequestType,
         self.input_socket.send_multipart(msg, copy=False)
 
     def add_request(self, request: EngineCoreRequest) -> None:
+        # NOTE: text prompt is not needed in the core engine as it has been
+        # tokenized.
+        request.prompt = None
         self._send_input(EngineCoreRequestType.ADD, request)
 
     def abort_requests(self, request_ids: List[str]) -> None:
@@ -257,6 +260,9 @@ async def _send_input(self, request_type: EngineCoreRequestType,
         await self.input_socket.send_multipart(msg, copy=False)
 
     async def add_request_async(self, request: EngineCoreRequest) -> None:
+        # NOTE: text prompt is not needed in the core engine as it has been
+        # tokenized.
+        request.prompt = None
         await self._send_input(EngineCoreRequestType.ADD, request)
 
     async def abort_requests_async(self, request_ids: List[str]) -> None: