do preprocessing properly

rubra-ai · Jun 12, 2024 · 49f4197 · 49f4197
1 parent fa74d02
commit 49f4197
Showing 1 changed file with 7 additions and 8 deletions.
diff --git a/vllm/entrypoints/openai/serving_chat.py b/vllm/entrypoints/openai/serving_chat.py
@@ -179,7 +179,6 @@ def _parse_chat_message_content(
     ) -> ChatMessageParseResult:
         role = message["role"]
         content = message.get("content")
-
         if content is None:
             return ChatMessageParseResult(messages=[], image_futures=[])
         if isinstance(content, str):
@@ -210,20 +209,20 @@ async def create_chat_completion(
         try:
             conversation: List[ConversationMessage] = []
             image_futures: List[Awaitable[ImagePixelData]] = []
-
-            raw_msgs = request.messages
-            if request.tools:
-                print("==================tools====================")
-                tools = [t.model_dump() for t in request.tools]
-                raw_msgs = preprocess_input(msgs=raw_msgs, tools=tools)
+            print("==================create chat completion====================")
 
             for msg in request.messages:
                 chat_parsed_result = self._parse_chat_message_content(msg)
 
                 conversation.extend(chat_parsed_result.messages)
                 image_futures.extend(chat_parsed_result.image_futures)
 
-            conversation = raw_msgs
+
+            if request.tools:
+                raw_msgs = request.messages
+                tools = [t.model_dump() for t in request.tools]
+                raw_msgs = preprocess_input(msgs=raw_msgs, tools=tools)
+                conversation = raw_msgs
 
             prompt = self.tokenizer.apply_chat_template(
                 conversation=conversation,