Update app.py

app.py

@@ -316,6 +316,7 @@ async def generate_openai_response(full_response: str, request_id: str, model: s
 
 
 async def stream_openai_response(response, request_id: str, model: str, token: str):
+    global TOKEN_INDEX
     """流式返回OpenAI API格式的响应"""
     timestamp = int(time.time())
     full_response = ""