Having beaten my head on this for a while (and shipped the first reasonably complete ML framework that runs on Metal) Apple's opinion as expressed by their priorities is that it's just not important.

> reliably run TensorFlow

What reliability issues are you having with TensorFlow on M1 Macs?

We've followed five different instructional and documentation pages to make it happen and none seem to consistently install. Throw in a corporate system where you need IT for root access to make changes and it is game over. So i've got an M1-max fully loaded and cant get TF running on it.

Now i've got a team of data scientists in a fully MBP shop and we're holding off upgrades to M1 until this all gets resolved.

On my personal M1, I managed to make it work, but its hard to know the layers of changes made and what exactly allowed it to work.

You can get off this GPU circus and simply go with purpose-built AI solutions.

You can buy single tensor accelerators from Google: https://www.coral.ai/products/

You can buy a bunch of those integrated into a single PCI-E card. https://iot.asus.com/products/AI-accelerator/AI-Accelerator-...

Cheap too. Some of these work with Mac. More of them work for PC, because the hardware interface is outside of Apple's thin vertical slice/garden.

These are devices for Tensorflow Lite which is more appropriate for IoT etc. not doing the intensive initial training of a complex model

Could be worth tracking what you did and make a new set of instructions, and trying to reproduce with a fresh install.

This is something Apple should pay people for.

Deep learning support for Mac is not going to happen at a level of quality you can rely on for research & dev work (like PyTorch + TensorFlow). The underlying problem is no big company cares about Mac platform and the work to maintain framework support for a specific piece of hardware is way beyond a hobby project. If you want your own on-prem hardware just buy Nvidia.