Enable integration with Metal API for accelerated generation on Apple Silicon

[AriaShishegaran]

Hi, thanks for the fantastic job. Do you plan to add integration with Apple Silicon, right now it is actively looking for CUDA cores and fails to do so, therefore, sticking with the CPU.

[aedocw]

There's a PR for that available here:
#181

Prior to that PR, I had done some testing and found it made no improvement. I think it will be useful if/when there's basically full CUDA compatibility and it work with deepspeed. I have not checked for progress in the last few weeks, but when I did last check I saw there were still quite a few missing operators that XTTS needs.

[AriaShishegaran]

Got it. is this, btw, something that needs to be addressed on Pytorch's side for it to fully leverage Metal API?