fix

app.py

@@ -220,12 +220,12 @@ with gr.Blocks() as demo:
         server_rtc_configuration=get_hf_turn_credentials(ttl=3_600 * 24 * 30)
     )
     with gr.Accordion(label="Additional Inputs"):
-        sp = gr.Textbox(label="System Prompt", value="You are a helpful assistant."),
+        sp = gr.Textbox(label="System Prompt", value="You are a helpful assistant.")
         slider = gr.Slider(label="Max New Tokens", minimum=100, maximum=2000, step=10, value=700)
         image = gr.Image()
 
     webrtc.stream(
-        ReplyOnPause(response),  # type: ignore
+        ReplyOnPause(generate),  # type: ignore
         inputs=[webrtc, chatbot, sp, slider, image],
         outputs=[chatbot],
         concurrency_limit=100,