enable xla fsdp

[hanwen-sun]

What does this PR do?

We have supported xla fsdp in accelerator, so we should enable xla fsdp in transformers.
the support of xla fsdp in accelerate: huggingface/accelerate#2176 and huggingface/accelerate#2941

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.

[amyeroberts]

cc @muellerzr @SunMarc

[muellerzr]

Much appreciated for the enable!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[amyeroberts]

I don't think we can do this directly. AFAICT this was enabled in 0.28 but the required version in our setup.py for accelerate is 0.21. Instead we'll need to add a condition checking the accelerate version is >= 0.28

[hanwen-sun]

I don't think we can do this directly. AFAICT this was enabled in 0.28 but the required version in our setup.py for accelerate is 0.21. Instead we'll need to add a condition checking the accelerate version is >= 0.28

So I need to add a condition checking that accelerate version is >= 0.28?

[amyeroberts]

@hanwen-sun Yes - you should add is_accelerate_available("0.28.0") to the if statement

[hanwen-sun]

@hanwen-sun Yes - you should add is_accelerate_available("0.28.0") to the if statement

thank you, I have updated.

[SunMarc]

Nice !

[amyeroberts]

Thanks!

[steveepreston]

Created #34197 to revert this change due to #34176 issue.