Update app.py

app.py

@@ -70,13 +70,15 @@ with gr.Blocks() as demo:
     min_new = gr.Slider(0, 512, value=160, step=8, label="min_new_tokens")
 
     chat = gr.ChatInterface(
-        fn=lambda m, h: chat_fn(m, h, system_box.value, temperature.value, top_p.value, max_new.value, min_new.value),
-        title=None,
-        additional_inputs=[system_box, temperature, top_p, max_new, min_new],
-        retry_btn="Regenerate",
-        undo_btn="Undo Last",
-        clear_btn="Clear",
-        queue=True,  # queue is recommended (and required for ZeroGPU concurrency)
+        fn=gradio_fn,
+        examples=[
+            "Hello!",
+            "How can I overcome fear of failure?",
+            "How do I forgive someone who hurt me deeply?",
+            "What can I do to stop overthinking?"
+        ],
+        chatbot=gr.Chatbot(elem_classes="chatbot"),
+        theme="compact",
     )
 
 if __name__ == "__main__":