It seems everyone wants deep learning to work on M1 chips. At this point, I am quite sure there is enough interest for major frameworks to consider supporting it.
You'd think Apple would be doing their best to help with this, considering they market their "neural engine" pretty heavily.