Update app.py

app.py

@@ -10,6 +10,8 @@ free CPU tier.
 Entry point for Hugging Face Spaces (Gradio SDK looks for app.py).
 """
 
+
+import spaces
 import os
 
 import gradio as gr
@@ -82,7 +84,7 @@ model = model.to(DEVICE)
 model.eval()
 print(f"Model loaded on {DEVICE}.")
 
-
+@spaces.GPU(duration=120)
 def _generate(text: str, max_new_tokens: int = 400) -> str:
     """Run one generation with the current adapter state."""
     messages = [