feat: expose the strict flag to allow catching missing model layers while loading a checkpoint

[HarikrishnanBalagopal]

What does this PR do?

Currently model loading succeeds even when a layer is missing from the saved model weights.
This pull reqeust exposes the strict flag from https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module.load_state_dict
to allow raising an error when the checkpoint is incomplete (missing some layers).

Also see pytorch/pytorch#82963

Fixes

  warnings.warn(
There were missing keys in the checkpoint model loaded: ['lm_head.weight'].

Related issues:

• bart-large-xsum model: There were missing keys in the checkpoint model loaded: ['model.encoder.embed_tokens.weight', 'model.decoder.embed_tokens.weight', 'lm_head.weight']. #29128
• T5 model: There were missing keys in the checkpoint model loaded: ['encoder.embed_tokens.weight', 'decoder.embed_tokens.weight', 'lm_head.weight']. #27972

Before submitting

• [n] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• [y] 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.

• trainer: @muellerzr and @SunMarc

[github-actions]

Hi 👋, thank you for opening this pull request! The pull request is converted to draft by default. When it is ready for review, please click the Ready for review button (at the bottom of the PR page).

[SunMarc]

Not sure we want to add this flag in trainer as this is related to model loading cc @muellerzr This is something that should be fixed when trying to load the model before the training.

[HarikrishnanBalagopal]

Not sure we want to add this flag in trainer as this is related to model loading cc @muellerzr This is something that should be fixed when trying to load the model before the training.

I understand the concern but the train function also has resume_from_checkpoint which does model loading.

[muellerzr]

This should be a training arg instead, so we don't need to change all of the function signatures. (I'm open to this being a thing since during resume we don't do that outside, however it should be in TrainingArguments instead)