Running a model's predictions on Apple's custom processors (M-series chips) rather than traditional CPUs or GPUs, enabling faster performance on Mac devices.