Skip to content

Add UNDIAL, a new unlearning algorithm #89

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 26 commits into from
May 22, 2025
Merged

Add UNDIAL, a new unlearning algorithm #89

merged 26 commits into from
May 22, 2025

Conversation

@dong-river
Copy link
Contributor

@dong-river dong-river commented Apr 1, 2025 •
edited by Dornavineeth
Loading

What does this PR do?

  • Added UNDIAL, a new unlearning algorithm publised in NAACL.
  • Tested on TOFU

Before submitting

  • This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
  • Have you gone through the contributions guide?
  • Are your changes documented? Read documentation guidelines here.

molereddy and others added 15 commits March 1, 2025 09:13
@molereddy
Fix hyperlinks in README (locuslab#2)
54d3560 
* testing commit

* Fixes

* cleanup
@Dornavineeth
Fixed DPO command
4c36e4f
@Dornavineeth
download idk
f7a69de
@Dornavineeth
Merge pull request locuslab#3 from Dornavineeth/dpo_fix
1bd4411 
Fix tofu_unlearn.sh for IdKDPO method.
@Dornavineeth
Revert "Dpo fix"
332af36
@Dornavineeth
Merge pull request locuslab#4 from Dornavineeth/revert-3-dpo_fix
7d6aef3 
Revert "Dpo fix"
@Dornavineeth
download idk data
f468efb
@Dornavineeth
fix dpo experiment config
ca8d503
@Dornavineeth
Merge pull request locuslab#5 from Dornavineeth/dpo_fix
dde60d3 
IdkDPO fix
@molereddy
Merge branch 'locuslab:main' into main
6367fb6
@Dornavineeth @molereddy
RMU (locuslab#6)
8b073d6 
* IdkDPO script fix in tofu_unlearn.sh (locuslab#65)

* Fix hyperlinks in README
* Download I don't know data in setup_data.py
* Fix tofu_unlearn.sh for IdkDPO

---------

Co-authored-by: Anmol Mekala <49127549+molereddy@users.noreply.github.com>

* overwrite=True

* RMU added

* Fix ref model device

* ruff fix

* RMU updated

* Update rmu.py

* Update README.md: add RMU

* Added references and renamed functions

---------

Co-authored-by: Anmol Mekala <49127549+molereddy@users.noreply.github.com>
@molereddy
Merge branch 'locuslab:main' into main
855c5f3
@Dornavineeth
Merge branch 'locuslab:main' into main
4fb577c
@Dornavineeth @molereddy
Add structure to contributions, setup leaderboard, update documentati…
dccb831 
…on (locuslab#8)

* docs: updates, small corrections, re-formats

* modified ruff commands

* modified ruff commands

* CI/CD minor updates

* added contributing + leaderboard

* fix minor spelling misatkes

* docs: bunch of minor updates

* docs fixes

---------

Co-authored-by: molereddy <m.anmolreddy@gmail.com>
@molereddy
Merge branch 'locuslab:main' into main
e3c4709
@molereddy
Copy link
Collaborator

Thank you for this!

Can you make these additions

  1. There are methods which run on top of unlearning loss functions (like SOUL, GRU (Add gru component #83). Since you are introducing a loss function, it would be great if you can make the undial_loss function available outside your trainer code so it can be imported in these methods as we begin to also support them.
  2. We've added guidelines on contributing to our repository, including sharing experimental settings, results, scripts, and to run. Could you review them and update this PR accordingly?

Summary of steps to contribute

  • Create a folder under community/methods/.
  • Include your method details, results, hyperparameter search scripts, and steps.
  • Update the leaderboard with your results.

We've provided a template here.

@dong-river
Copy link
Contributor Author

@molereddy Thank you for the suggestions. I fixed the part I could understand. And send a private message about things I don't understand. I hope at least what I did in this PR is able to be merged. Let me know if there's any fatal issues. Thank you!

@molereddy
Copy link
Collaborator

molereddy commented Apr 13, 2025 •
edited
Loading

I've merged our upstream changes onto your branch. If you can pull, test the code and add some results, that would be great. Else mark PR for review: we can then just test on our end and merge the code in in the coming days.

@dong-river dong-river marked this pull request as ready for review April 25, 2025 20:49
@dong-river
Copy link
Contributor Author

Sorry for the delay. I had a bit trouble running my method on 7B which requires LoRA implementation as my method need to run TWO 7B model in parallel. It would be great if LoRA could be supported.

I would really appreciate if you could test my methods like all other using the same standard and follow the hyperparam search in my description in the PR (which should not be hard given you have the pipeline ready). I also turned the PR for review.

@Dornavineeth Dornavineeth self-requested a review May 4, 2025 04:17
Dornavineeth
Dornavineeth approved these changes May 22, 2025
View reviewed changes
Copy link
Collaborator

@Dornavineeth Dornavineeth left a comment

Choose a reason for hiding this comment

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

Great Work.

Currently it is tested only on TOFU 1B model.

@Dornavineeth Dornavineeth merged commit 0d35d15 into locuslab:main May 22, 2025
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@Dornavineeth Dornavineeth Dornavineeth approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@dong-river @molereddy @Dornavineeth