[bugfix] fix megatron opsd with pp

[hjh0119]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello, 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 addresses a bug in the Megatron OPSD (One-Pass Self-Distillation) trainer, specifically concerning its interaction with pipeline parallelism. The changes streamline the process of computing and propagating teacher logits and labels, particularly for self-distillation scenarios. By centralizing teacher model setup and logit computation, and ensuring opsd_teacher_labels are correctly managed, the fix resolves inconsistencies and redundancies in the training pipeline.

Highlights

• Refactored Teacher Logit Computation: The logic for determining the teacher model and its associated context (e.g., disabling adapters) has been centralized and moved outside the batch processing loop within _compute_teacher_logits_local.
• Improved OPSD Label Handling: The opsd_teacher_labels are now explicitly extracted and stored in the encoded_batch for both local and API-based teacher logit computations, ensuring correct propagation.
• Eliminated Redundant Computation: Redundant teacher logit computation within the forward_step method for self-distillation has been removed, as this process is now handled upstream in _compute_teacher_logits_local.
• Consistent Teacher Logit Execution: The conditional check that previously prevented _compute_teacher_logits from being called during self-distillation has been removed, ensuring consistent execution across all distillation modes.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The 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 /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

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 .gemini/ folder in the base of the repository. Detailed instructions can be found here.

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.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request refactors the teacher logit computation to better support pipeline parallelism, especially for on-policy self-distillation (OPSD). The changes centralize the logic and fix an issue where teacher logits were incorrectly computed per-pipeline-stage. However, the refactoring introduces a critical bug in the code path for fetching teacher logits from an API, where labels for OPSD are not moved to the correct device, which would cause a runtime error. My review provides comments and suggestions to fix this issue while retaining the benefits of the refactoring.