DPO trainer supports num_logits_to_keep to save memory

[xyangk]

What does this PR do?

Since transformers v4.45.0 supports num_logits_to_keep in the forward pass(check this PR), it can be useful for saving memory and speeding up training by computing logits only for the last completion tokens, especially in scenarios with very long prompts where labels are ignored (-100).

I implemented the logic to calculate num_logits_to_keep based on the labels and tested this on DPO, resulting in substantial memory savings.

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? 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.
• 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.

[kashif]

@xyangk thank you! can you run the pre-commit task via the makefile to fix the formatting?

[xyangk]

Sure!

[xyangk]

Implemented the calculation of num_logits_to_keep on the new master code, fixed a bug in the previous calculation, and used more rigorous test code.

[xyangk]

Hi @kashif , I wanted to check in on this PR. Please let me know if there's anything further I can clarify or update. Thank you for your time and review!

[kashif]

@xyangk can you use the transformers helper check_min_version instead of implementing the logic yourself?

[xyangk]

@xyangk can you use the transformers helper check_min_version instead of implementing the logic yourself?

I've implemented the suggested changes. Thank you for the feedback!

[kashif]

@xyangk in view of the fix #2332 can you kindly confirm that this feature is not affected by this?

[xyangk]

in view of the fix #2332 can you kindly confirm that this feature is not affected by this?

I've reviewed the code in realtion to fix #2332. I can confirm that this feature remains unaffected by those changes. I've also added some comments to improve clarity.

[qgallouedec]

You don't need it

https://github.com/huggingface/trl/blob/main/setup.py#L82

[xyangk]

Sorry I missed that the transformers version was already updated. Since the current PR is already merged, I'll create a new PR to remove the version check. Thank you for cathing this!

[qgallouedec]

#2343