Arm64 macOS _coreML and nn

Dear PyTorch team,

I am wondering the newest 1.12.1 for apple m1 devices, to_device(“cuda”) is not working but I say that the torch.backend properties has _coreML and nn specifically for apple neural engine. I am wonder when I want to use it on apple M1 pro/MAX gpu, what should I do.

Thanks,

Jianshu