[WIP][RL] fix fp8 update weight

[zhuzilin]

Motivation

Modifications

For RL training, we need to call the weight_loader of the params multiple times for weight updation, while the current implementation will discard the weight_loader relevant attr (e.g. load_merged_column_weight).

The current fix is a super hacky one. I'm not sure how to correctly fix this...

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.

[gemini-code-assist]

Summary of Changes

Hello @zhuzilin, 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 provides a temporary fix for an issue in FP8 weight updates during RL training. The problem stemmed from the loss of critical weight_loader attributes when weight parameters were re-assigned. The current solution prevents this re-assignment, ensuring that the necessary parameter metadata is preserved across updates, though the author notes it is a 'hacky' solution.

Highlights

• FP8 Weight Update Issue: This pull request addresses a problem in FP8 weight updates, particularly relevant for Reinforcement Learning (RL) training, where attributes of the weight_loader (e.g., load_merged_column_weight) were being discarded.
• Preservation of Parameter Attributes: The core change involves commenting out lines that re-instantiated layer.weight and layer.weight_scale_inv as new torch.nn.Parameter objects. This prevents the loss of existing attributes associated with these parameters, which was occurring when they were re-assigned.

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 is currently in preview and 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 to provide feedback.

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

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

The pull request fixes an issue where the weight_loader attribute was being discarded, preventing multiple weight updates in RL training. The fix involves commenting out the lines that reassign the weight and weight_scale_inv as torch.nn.Parameters. The author acknowledges that this is a temporary solution and seeks a more correct fix.

[gemini-code-assist]

Consider removing these commented-out lines to reduce code clutter and improve readability. If the code is intentionally commented out for debugging or future use, add a comment explaining the reason.