A method of optimizing a Checkpoint model to increase the speed of inference, while reducing file size and VRAM cost.