Update app.py

app.py

@@ -10,13 +10,21 @@ def respond(message, history):
         messages.extend(history)
         
     messages.append({"role" : "user", "content" : message})
-    response = client.chat_completion(
+    response = ""
+    
+    client.chat_completion(
         messages, 
         max_tokens = 100, #change length
         temperature = 0, #randomness
         top_p = 0.3
+        stream = True
     )
 
+    for chats in client.chat_completion:
+        token = message.choices[0].delta.content
+        response += token
+        yield response
+
     print(response["choices"][0]["message"]["content"].strip())
     return response["choices"][0]["message"]["content"].strip()