Update main.py

main.py

@@ -34,8 +34,8 @@ class ChatCompletionRequest(BaseModel):
 @app.get("/stream")
 async def chat(prompt = "Once upon a time there was a "):
     completion = llm(prompt)
-    yield prompt
     async def server_sent_events(chat_chunks):
+        yield prompt
         for chat_chunk in chat_chunks:
             yield chat_chunk
         yield "[DONE]"