Skip to content

HF Causal Reward Model #6

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.

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

Closed
wants to merge 7 commits into from
Closed

HF Causal Reward Model #6

wants to merge 7 commits into from

Conversation

bcui-db
Copy link
Collaborator

@bcui-db bcui-db commented Feb 16, 2025

No description provided.

@alextrott16 alextrott16 mentioned this pull request Feb 25, 2025
bcui-db added a commit that referenced this pull request Jun 17, 2025
Adds a "causal" version of the classifier RM. Causal means that the
classifier RM uses the same architecture as the causal base model,
devoting the EOS token's logit to be the classifier logit.

Note that [this PR](#6)
introduces causal preference RMs. The current PR is based off of that
implementation, so it may make sense to wait on that PR or have this PR
merge into that branch.

This has been used extensively.
Here is an
[mlflow](https://dbc-559ffd80-2bfc.cloud.databricks.com/ml/experiments/4128924280951945/runs/d9f3526199b9402c8275786fb5b0ec14/model-metrics?o=7395834863327820)
run using the latest code. Notice how it runs.

---------

Co-authored-by: Brandon <[email protected]>
Co-authored-by: bcui-db <[email protected]>
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.

3 participants