add set_epoch for MpDeviceLoaderWrapper

[hanwen-sun]

What does this PR do?

in transformers' trainer.py, when using xla resume, the data_iterator used during resumption is different from the original data_iterator. It is necessary to use set_epoch with the current epoch value: https://github.com/huggingface/transformers/blob/3bf6dd8aa141ff6bebd579649e9d0ba8bf5d03db/src/transformers/trainer.py#L2264. However accelerator adds an additional wrapper for xla: MpDeviceLoaderWrapper, which does not have a set_epoch method, this leads to incorrect epoch value and data being used in subsequent training (the random seed in the data loader's sampler is related to the epoch value). To resolve this issue, we add a set_epoch method to MpDeviceLoaderWrapper.

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.

[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.

[BenjaminBossan]

@hanwen-sun Thanks for the PR. The function body is over-indented by one space, could you please fix that?

[hanwen-sun]

@hanwen-sun Thanks for the PR. The function body is over-indented by one space, could you please fix that?

sure, I have fixed that

[muellerzr]

Thanks for adding this!