fix bug

app.py

@@ -40,15 +40,16 @@ def respond(
     # )
     # return response["choices"][0]["message"]["content"]
     response = ""
-
-    for message in llm.create_chat_completion(
+    completion = llm.create_chat_completion(
         messages,
         max_tokens=max_tokens,
         stream=True,
         temperature=temperature,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+    )
+
+    for message in completion:
+        token = message['choices'][0]['delta']['content']
 
         response += token
         yield response