Update app.py

app.py

@@ -210,7 +210,7 @@ EXAMPLES = [
 
 
 # Definir la función simple_chat
-@spaces.GPU()
+# @spaces.GPU()
 async def simple_chat(message: dict, temperature: float = 0.8, max_length: int = 4096, top_p: float = 1, top_k: int = 10, penalty: float = 1.0):
     # Cargar el modelo preentrenado
     model = AutoModelForCausalLM.from_pretrained(
@@ -220,6 +220,8 @@ async def simple_chat(message: dict, temperature: float = 0.8, max_length: int =
         trust_remote_code=True
     )
 
+    model.to('cuda')
+
     conversation = []
 
     if "file" in message and message["file"]: