Add tensor.detach() back to update weight util #8691
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @hebiao064, 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!
I've addressed a serialization error that was occurring during weight updates, particularly in scenarios involving Megatron Resharding to SGLang. The core of the fix involves re-introducing a tensor.detach() call in the weight synchronization utility. This ensures that tensors are properly prepared before serialization, resolving the compatibility issue that arose from a previously missed detail during code migration.
Highlights
- Serialization Fix: I've added
tensor.detach()to the_preprocess_tensor_for_update_weightsfunction call within theupdate_weightsutility. This change is critical for preventing serialization errors, especially when using Megatron Resharding with SGLang.
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 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 or fill out our survey 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
-
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. ↩
There was a problem hiding this comment.
Code Review
This pull request addresses a potential serialization error by adding .detach() to the tensor before it's passed to the serialization utility. This is a correct and necessary change to ensure that tensors requiring gradients are handled properly during weight updates, preventing the serialization of the entire computation graph. The change is well-contained and directly resolves the issue described. I have reviewed the changes and find them to be correct and have no further suggestions for improvement.
Motivation
Add tensor.detach() to update weight util to avoid serialization error for Megatron Resharding to SGLang
Context
When I copy the update weight from VERL Code to SGLang code: #8267, I missed the
tensor.detach()which is only needed in VERL Megatron TP=1 case.https://github.com/volcengine/verl/pull/2720/files#diff-3ea40b07302076f020ceb946048c8129b9bd9141feebef80312da55c431de3feL161
Hence we need to add it back for compatibility.
Modifications
Accuracy Test
Benchmark & Profiling
Checklist