Update app.py

app.py

@@ -6,6 +6,7 @@ from transformers import (
     TextIteratorStreamer,
 )
 import gradio as gr
+import spaces
 
 # Load model and tokenizer
 model_id = "microsoft/bitnet-b1.58-2B-4T"
@@ -17,6 +18,7 @@ model = AutoModelForCausalLM.from_pretrained(
     device_map="auto"
 )
 
+@spaces.GPU
 def respond(
     message: str,
     history: list[tuple[str, str]],