Skip to content

Fix DPO unit test fail and refactor#554

Merged
austin362667 merged 4 commits into
mainfrom
tcc/dpo-ci-hotfix
Feb 9, 2025
Merged

Fix DPO unit test fail and refactor#554
austin362667 merged 4 commits into
mainfrom
tcc/dpo-ci-hotfix

Conversation

@Tcc0403

@Tcc0403 Tcc0403 commented Feb 1, 2025

Copy link
Copy Markdown
Collaborator

Summary

#521 introduced a unit test failure caused by the numerical issue in bf16.

This PR relaxes the atol for chosen_rewards and rejected_rewards only.
Note: It's just a temporary fix for ci.

This PR also removes duplicated calculations on both terms.

Testing Done

  • Hardware Type:
  • run make test to ensure correctness
  • run make checkstyle to ensure code style
  • run make test-convergence to ensure convergence

@kashif

kashif commented Feb 2, 2025

Copy link
Copy Markdown
Contributor

thanks @Tcc0403

@shivam15s

Copy link
Copy Markdown
Collaborator

Can you raise this PR in LinkedIn?

@Tcc0403

Tcc0403 commented Feb 7, 2025

Copy link
Copy Markdown
Collaborator Author

@shivam15s do you mean linkedin branch?

@shivam15s

Copy link
Copy Markdown
Collaborator

yep

@austin362667 austin362667 left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks! 🚀

@Tcc0403

Tcc0403 commented Feb 7, 2025

Copy link
Copy Markdown
Collaborator Author

@shivam15s It's on linkedin repo's branch!

@austin362667 austin362667 merged commit 4dccc26 into main Feb 9, 2025
@austin362667 austin362667 deleted the tcc/dpo-ci-hotfix branch February 9, 2025 14:52
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants