Do any deep learning frameworks even have metal support? I could see inference working well on these laptops but they lack so much of the specialized hardware I'd be surprised if training was even possible for most useable models.

Apple is maintaining a Tensorflow plugin: https://developer.apple.com/metal/tensorflow-plugin/

There's been work on PyTorch to move some over, but it's still all CPU (which still, on an M1 isn't HORRIBLE) last time I checked.

Tensorflow supports Metal AFAIK.

Wouldn't you use the Apple Neural Engine for that?

https://www.infoq.com/news/2020/11/apple-tensorflow-accelera...

"Recently Apple released the new M1 "system on a chip," which not only contains a built-in GPU, but also includes a 16-core "Neural Engine" which supports 11 trillion operations per second. Apple claims the Neural Engine will support up to 15x improvement in ML computation."

The neural engine seems to be only about inference. For training it seems most systems rely on Metal like the Apple Tensorflow plugin [1]. But I have never tried to do ML on macs so I am maybe wrong.

[1] https://developer.apple.com/metal/tensorflow-plugin/

Do people really train locally? I'd have thought the field moved to aws or some other hpc by now. Seems like a waste buying such a nice laptop to just melt and abuse it when you can abuse amazon's hardware instead. The battery won't be happy being discharged every day since it seems like in my experience macbooks don't bypass the battery when on ac power unless you start the computer up with the batty unplugged (not so easy on newer computers)