Update app.py

app.py

@@ -60,7 +60,7 @@ device_map = infer_auto_device_map(model, max_memory={0: "80GB", 1: "80GB", "cpu
 
 # Load the model with the inferred device map
 # model = load_checkpoint_and_dispatch(model, model_id, device_map=device_map, no_split_module_classes=["GPTJBlock"])
-# model.half()
+model.half()