After saving and loading a machine, it runs inference on GPU (specifically, for XGBoost), is there an interface to move the model back to GPU?