Fix KeyError in extract_model_from_parallel for partial torch.compile

[amanzoni1]

What does this PR do?

This PR fixes a KeyError: '_orig_mod' crash within the extract_model_from_parallel utility (called by accelerator.unwrap_model and subsequently by the transformers.Trainer) when attempting to unwrap a model that contains compiled submodules but is not compiled at the top level.

The original code assumed that if has_compiled_regions(model) returned True, the top-level model object must necessarily have the _orig_mod attribute (or be a wrapper that exposes it via __dict__), leading to the crash when the model was only partially compiled.

The fix introduces a safe check (if "_orig_mod" in model.__dict__:) before attempting to access the attribute in the elif has_compiled: block.

Fixes #3879

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• [ x ] Did you read the contributor guideline,
  Pull Request section?
• [ x ] 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.
• [ x ] Did you write any new necessary tests?

Who can review?

@SunMarc
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.

[SunMarc]

Thanks for this ! Would that solve your issue @Giuseppe5 ?

[Giuseppe5]

I believe so, I'm going to test it locally and let you know if something else breaks.

Thanks a lot @amanzoni1 and @SunMarc!

[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.