Update app.py

app.py

@@ -150,6 +150,10 @@ def messages():
                     if chunk:
                         response_text.append(chunk)
 
+                # 检查是否是最终响应
+                if data.get('final', False):
+                    response_event.set()
+
             def on_query_complete(data):
                 response_event.set()
 
@@ -198,7 +202,7 @@ def messages():
                 "delta": {"stop_reason": "end_turn", "stop_sequence": None},
                 "usage": {"output_tokens": len(''.join(response_text))},
             })
-            yield create_event("message_stop", {"type": "message_stop"})
+            yield create_event("message_stop", {"type": "message_stop"})  # 确保发送 message_stop 事件
             log_request(request.remote_addr, request.path, 200)
 
         return Response(generate(), content_type='text/event-stream')