Update app.py

app.py

@@ -41,7 +41,7 @@ def generate_response():
     messages = messages[1:]
 
     payload = json.dumps({
-        'model': 'NBG-plus',
+        'model': 'nanbeige2-turbo-0611',
         'messages': messages,
         'max_tokens': 4096,
         'temperature': 0.7,
@@ -65,7 +65,7 @@ def stream_response(resp):
         if len(line) == 0:
             continue
         r = json.loads(line[6:])
-        content = r['modelServerData']['choices'][0]['delta']['content']
+        content = r['choices'][0]['delta']['content']
         yield content[last_length:]
         last_length = len(content)