Update app.py

app.py

@@ -81,12 +81,13 @@ with gr.Blocks() as demo:
     # Automatically handle submission on Enter key press with streaming
     def handle_submit(message, history, system_message, max_tokens, temperature, top_p):
         history.append((message, None))  # Append user's message first
-        chatbot.update(history)  # Display user's message before response
+        
+        # Stream the assistant's response and update the history
         for response in respond(message, history, system_message, max_tokens, temperature, top_p):
-            history[-1] = (message, response)  # Update the last response with streaming content
-            chatbot.update(history)
-        return history, ""
+            history[-1] = (message, response)
+            yield history, ""  # Yield updated history to display in the chatbox
 
+    # Use the new Gradio `chatbot.update` pattern by returning the updated value
     msg.submit(
         handle_submit,
         inputs=[msg, chatbot, system_msg, max_tokens, temperature, top_p],