Implement On-Policy Distillation #444
Merged
Conversation
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
Contributor
Summary of ChangesHello @garyzhang99, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request integrates On-Policy Distillation into the "trinity" framework, drawing inspiration from the Tinker library. The changes enable a student model to learn from a teacher model by comparing their log probabilities on sampled trajectories. This is achieved through the addition of a specialized advantage function, a pure importance sampling policy loss, and a new workflow that orchestrates the data collection and processing, including the storage of teacher-generated log probabilities within the experience objects. Highlights
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request introduces the implementation of On-Policy Distillation, a technique for improving a student model with guidance from a teacher model. The changes are well-structured, adding a new algorithm type, advantage function, policy loss, and workflow. The implementation correctly follows the principles of on-policy distillation. I've identified a potential correctness issue in the workflow's padding logic and a minor opportunity for code simplification. Overall, this is a solid contribution that adds a valuable feature.
garyzhang99
changed the title
hiyuchang
reviewed
Dec 18, 2025
hiyuchang
reviewed
Dec 18, 2025
Collaborator
Author
|
/unittest-all |
Summary
Failed Tests
Skipped
Tests
Github Test Reporter by CTRF 💚 |
Collaborator
Author
|
/unittest-all |
Summary
Skipped
Tests
Github Test Reporter by CTRF 💚 |
pan-x-c
reviewed
Dec 19, 2025
pan-x-c
reviewed
Dec 19, 2025
added 2 commits
December 19, 2025 14:30
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Description
As the title says.
Reference: https://thinkingmachines.ai/blog/on-policy-distillation/
We run a concrete example as below:
Refer to the readme.md in example for more details.
Checklist
Please check the following items before code is ready to be reviewed.