You wouldnt be using coreml since thats apples software, but likely vulkan versions of llama.cpp or pytorch or what have you. How well that works at the moment i dont know