-
Notifications
You must be signed in to change notification settings - Fork 308
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
[Feature, NOMERGE] RLHF Rollouts (reopened) #1329
Conversation
@tcbegley I think we should merge/rebase main into this |
897649b
to
ebed304
Compare
I've rebased and fixed up the tests. There's still the changes from #1328 visible in the history here because I need the reward model for the rollout. Once that's merged (I think it's ready), I'll do a final merge / rebase and we should be ready here too. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I can't approve it bc it's my PR but LGTM anyway!
EOS_TOKEN_ID = 50256 | ||
|
||
def __init__( | ||
self, model, ref_model, reward_model, max_new_tokens=50, score_clip=10.0 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
long term we may want to make this reward model optional and leave to the user the option of computing reward at a different time (eg when populating the replay buffer)
Co-authored-by: Alessandro Pietro Bardelli <apbard@users.noreply.github.com>
Co-authored-by: Vincent Moens <vincentmoens@gmail.com>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! thanks
See #1315 for discussion