Add Question 189: Compute Direct Preference Optimization Loss#583
Open
zhenhuan-yang wants to merge 1 commit intoOpen-Deep-ML:mainfrom
Open
Add Question 189: Compute Direct Preference Optimization Loss#583zhenhuan-yang wants to merge 1 commit intoOpen-Deep-ML:mainfrom
zhenhuan-yang wants to merge 1 commit intoOpen-Deep-ML:mainfrom
Conversation
moe18
reviewed
Mar 5, 2026
|
|
||
| # Compute loss using log-sigmoid for numerical stability | ||
| # Loss = -log(sigmoid(logits)) = log(1 + exp(-logits)) | ||
| losses = np.log1p(np.exp(-logits)) |
Collaborator
There was a problem hiding this comment.
This overflows to inf when logits is large and negative
losses = np.logaddexp(0, -logits)
this might be more stable
Collaborator
There was a problem hiding this comment.
I would also add a test case to exploit this
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 adds a new medium-difficulty Deep Learning question on computing Direct Preference Optimization (DPO) loss for language model alignment.
Question Details
Implementation
np.log1pValidation
Educational Value
Covers an important modern technique for LLM alignment that's simpler and more stable than traditional RLHF, making it highly relevant for current ML practitioners.