Fix TPU checkpointing inside Trainer

[shub-kris]

What does this PR do?

This PR fixes the following issue:

Whenever save_steps is a multiple of logging_steps this step if self.control.should_log and self.state.global_step > self._globalstep_last_logged is invoked which calls the xm.mark_step and the checkpointing defined inside save_tpu here works fine.

 def _maybe_log_save_evaluate(self, tr_loss, grad_norm, model, trial, epoch, ignore_keys_for_eval):
        if self.control.should_log and self.state.global_step > self._globalstep_last_logged:
            if is_torch_xla_available():
                xm.mark_step()

but the execution halts when save_steps is not a multiple of logging_steps as the xm.mark_step is never called.

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?

@muellerzr @pacman100

[muellerzr]

Excellent investigative work :)

[pacman100]

Thank you @shub-kris for the quick fix and deep dive!

[shub-kris]

The issue was also discussed here

[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]

Thanks!