tensorflow_macos VS cross

https://github.com/apple/tensorflow_macos is no longer needed

Tensorflow has a CoreML enabled version which run on ANE.

https://github.com/apple/tensorflow_macos

Apple released a guide on how to use the M1's integrated Neural Chip in TensorFlow. Have a look at this Apple documentation page (and maybe also this GitHub that talks about TensorFlow together with Apple's own ML Compute platform).

Tensorflow does work on Apple Silicon

Apple did some work to optimize tensorflow for M1, can be found here https://github.com/apple/tensorflow_macos It's alpha, but works fine, I tried it

https://github.com/apple/tensorflow_macos/issues/25

https://forums.macrumors.com/threads/apple-silicon-deep-lear...

It is expected that the M1 Max should have similar performance to a RTX-2080 or Titan X.

In case anyone is interested, in ran a fairly simple MNIST benchmark (proposed here : https://github.com/apple/tensorflow_macos/issues/25) on my recently acquired M1 Pro MBP (16-core GPU, 16GB RAM).

The only method I know of to download tensorflow on M1 macs is the one documented here: https://github.com/apple/tensorflow_macos

To compile a program with musl on a glibc system you can use cross-rs!

I cross compile to Mac, bsd, windows, etc cross ... Works great for me with either docker or podman.

If docker is available for you, https://github.com/cross-rs/cross is another and reliable way to solve this kind of problem. I do use it regularly.

We are using https://github.com/cross-rs/cross.

There is some built-in support in rustc for cross-compiling, but getting the build to actually work can be tricky due to the need for an appropriate linker. Instead, we’re going to use the Cross crate, which used to be maintained by the Rust Embedded Working Group Tools group.

I have used cross before to cross compile from Linux to other Linux. It has a section on it's wiki about this. Maybe that could be of help.

You can use cross to build your application and copy the artifacts into an alpine armv7 container. It would also build faster due to using cross compilation rather than QEMU.

Looks like cross is the easiest way to get something cross-compiled but its Mac support is blocked behind building your own build image. Even that repo says that it might be broken.

Bonus: if you’re working with Rust and doing a lot of cross platform stuff, check out cross. It runs QEMU in docker so you can run tests on a bunch of different emulated targets easily- literally a one line setup, it’s kind of magical.

It's also not as naturally cross-compilable as Go, though that's partly a side-effect of not accepting being a semi-closed ecosystem to achieve that and cross exists as a stop-gap while things like cargo-zigbuild explore less drastic options.