Update app.py

app.py

@@ -554,14 +554,14 @@ def generate(message, temperature):
     stream = client.chat.completions.create(
         model="H4/zephyr-7b-beta",
         messages=message,
-        stream=False,
+        stream=True,
         max_tokens=1024,
         stop=["output\n"],
         temperature=temperature,
         #timeout=30,
     )
 
-    response = stream.output
+    response = stream.text
 
     # The reason why the library method is not used here is that if an error occurs,
     #    the returned data will not be a stream, and using the official library will result in an error.