support for XLA on metal/mps (tensorflow, jax, pytorch)

I am curious about XLA support. Is this planned for tensorflow-metal/tensorflow-macos?

That would likely work seamlessly for JAX and PyTorch as well.

Up vote post of ngam Down vote post of ngam
195 views
Posted by
ngam
Reply
Add a Comment

Accepted Reply

Hi @ngam!

Unfortunately we can't comment here any future plans we have. But I will pass forward that there is interest for support for XLA in the developer community.

Posted by
Frameworks Engineer
Up vote reply of Frameworks Engineer Down vote reply of Frameworks Engineer
Post marked as solved
Add a Comment

Replies

Hi @ngam!

Unfortunately we can't comment here any future plans we have. But I will pass forward that there is interest for support for XLA in the developer community.

Posted by
Frameworks Engineer
Up vote reply of Frameworks Engineer Down vote reply of Frameworks Engineer
Post marked as solved
Add a Comment

Thank you. I understand.

If there is a need for a lengthier explanation, I would be happy to provide one. Both TensorFlow and JAX will likely gain a lot in terms of performance if XLA is supported here. This will likely provide developers (and users) exciting tools to develop their models. It goes without saying that if that could be open-sourced, it will be great. :)

Posted by
ngam
Up vote reply of ngam Down vote reply of ngam
Add a Comment