Update app.py

app.py

@@ -27,6 +27,7 @@ model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype=torch.bfloat1
 # -------------------------------------------------
 # Helpers
 # -------------------------------------------------
+
 def build_messages(history, enable_thinking: bool):
     """Convert Gradio history to the chat template."""
     messages = []
@@ -90,7 +91,7 @@ with gr.Blocks(title="SmolLM3-3B Chat") as demo:
     msg.submit(
         user_fn, [msg, chatbot], [msg, chatbot], queue=False
     ).then(
-        chat_fn, [chatbot, enable_think, temperature, top_p, top_k, repetition_penalty], chatbot
+        chat_fn, [chatbot, enable_think, temperature, top_p, top_k, repetition_penalty, max_new_tokens], chatbot
     )
     clear.click(lambda: None, None, chatbot, queue=False)