Enables torchrun for XLA-based accelerators

[jeffhataws]

What does this PR do?

This PR enables torchrun for XLA-based accelerators (TPU/NeuronCore) by using torch.distributed XLA backend. It is dependent on the torch/xla change pytorch/xla#3609.

Example application is the AWS Neuron tutorial with HF Trainer that uses torchrun:

https://awsdocs-neuron.readthedocs-hosted.com/en/latest/frameworks/torch/torch-neuronx/tutorials/training/finetune_hftrainer.html

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@sgugger

[sgugger]

Thanks for your PR! There are a few problems with it though, as it breaks current support for PyTorch XLA.

[sgugger]

No the model needs to be wrapped in a DistributedDataParallel for TPU,

[jeffhataws]

Thanks. Limited the case to torchrun in the latest commit and clarified that currently DDP doesn't work with torch.distributed XLA backend yet (pytorch/pytorch#79164)

[sgugger]

In this case we should wait until it's supported. This will break the current TPU support.

[sgugger]

This should be moved to the _setup_devices method of TrainingArguments. It shouldn't be done on import.

[jeffhataws]

Thanks. Unfortunately I can't move it into _setup_devices for some reason as it run into device not found error. Will investigate.

[sgugger]

We don't have any runner testing on XLA for now, so this test will never be run FYI.

[jeffhataws]

Thanks. Will get help with running these tests.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.