Update main.py

main.py

@@ -62,7 +62,8 @@ async def completion(request: ChatCompletionRequest, response_mode=None):
 
 @app.post("/v1/chat/completions")
 async def chat(request: ChatCompletionRequest):
-    tokens = llm.tokenize([message.content for message in request.messages])
+    combined_messages = ' '.join([message.content for message in request.messages])
+    tokens = llm.tokenize(combined_messages)
     
     try:
         chat_chunks = llm.generate(tokens, max_tokens=request.max_tokens)