Update app.py

app.py

@@ -2,6 +2,7 @@ import torch
 
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 import gradio as gr
+import spaces
 from threading import Thread
 
 MODEL = "tiiuae/Falcon3-7B-Instruct-1.58bit"
@@ -36,6 +37,7 @@ model = AutoModelForCausalLM.from_pretrained(
 
 model = torch.compile(model)
 
+@spaces.GPU
 def stream_chat(
     message: str, 
     history: list,