Update app.py

app.py

@@ -2,6 +2,7 @@ import os
 import gradio as gr
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
+import spaces
 
 model = Llama(
     model_path=hf_hub_download(
@@ -31,6 +32,7 @@ def llama_o1_template(data):
     text = template.format(content=data)
     return text
 
+@spaces.GPU(duration=200)
 def generate_text(message, history, max_tokens=512, temperature=0.9, top_p=0.95):
     temp = ""
     input_texts = [llama_o1_template(message)]