Update app.py

app.py

@@ -6,7 +6,7 @@ import os
 
 @spaces.GPU
 def load_model(model_name):
-    return pipeline("text-generation", model=model_name, device="cuda", torch_dtype=torch.bfloat16, trust_remote_code=True, token=os.environ["token"])
+    return pipeline("text-generation", model=model_name, device_map="auto", torch_dtype=torch.bfloat16, trust_remote_code=True, token=os.environ["token"])
 @spaces.GPU
 def generate(
     model_name,