Refactor LigerFusedLinearPreferenceBase
#381
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
This was quick. Great refactor! |
shivam15s
previously approved these changes
Nov 14, 2024
Comment on lines
47
to
50
| _input, | ||
| weight, | ||
| target, | ||
| bias, |
There was a problem hiding this comment.
nit: I wrongly assumed that forward just takes positional arguments.
Could you also make input, weight target, bias keyword args?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
This PR refactors the
LigerFusedLinearPreferenceBaseclass to contain an abstractmethod corresponding to the calculation of the loss that needs to be implemented by all sub-classes.It also adds a new function to the class called
_compute_losswhich is mostly the same as the_compute_orpo_lossfunction introduced in #362 but makes it generic to calculate the NLL/Cross Entropy Loss plus accepts a custom loss function that implements a new alignment loss function.Most RLHF/RLAIF/Alignment algorithms state their final loss as
NLL + Beta * (Alignment_Loss)so adding the NLL logic inside the base class reduces repeated code.The _compute_loss function accepts
Testing Done
On A100-80G-SXM
make testto ensure correctnessmake checkstyleto ensure code stylemake test-convergenceto ensure convergence