Skip to content

fix post_residual_addition more generally#17286

Merged
Kangyan-Zhou merged 5 commits intosgl-project:mainfrom
nanjiangwill:fix/layernorm
Jan 23, 2026
Merged

fix post_residual_addition more generally#17286
Kangyan-Zhou merged 5 commits intosgl-project:mainfrom
nanjiangwill:fix/layernorm

Conversation

@nanjiangwill
Copy link
Copy Markdown
Contributor

@nanjiangwill nanjiangwill commented Jan 18, 2026

Motivation

fix post_residual_addition addition for non-GPU backend

#16561 only fixed cuda backend

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

Review Process

  1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
  2. Get approvals from CODEOWNERS and other reviewers.
  3. Trigger CI tests with comments or contact authorized users to do so.
    • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
  4. After green CI and required approvals, ask Merge Oncalls to merge.

@gemini-code-assist
Copy link
Copy Markdown
Contributor

Summary of Changes

Hello @nanjiangwill, 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 an issue where post_residual_addition was not being correctly applied in certain RMSNorm layer implementations, particularly for non-GPU backends. The changes ensure that this additional residual term is properly incorporated into the main residual path before the normalization step, thereby improving the accuracy and robustness of the RMSNorm layer across all supported hardware configurations and model types, including Gemma.

Highlights

  • Generalized Residual Addition: The pull request generalizes the handling of post_residual_addition to ensure it is correctly applied across various backend implementations of the RMSNorm layer.
  • Cross-Backend Compatibility: The post_residual_addition logic has been consistently integrated into forward_npu, forward_aiter, forward_hip, forward_cpu, forward_xpu, and forward_with_allreduce_fusion methods, ensuring proper behavior on non-GPU backends.
  • Gemma Model Support: The fix also extends to Gemma-specific RMSNorm implementations, including _forward_impl, forward_native, forward_cpu, and forward_npu for Gemma, ensuring consistent residual addition.

🧠 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.

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.

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request adds support for post_residual_addition to various normalization forward passes for non-GPU backends, ensuring consistent behavior across different hardware platforms. The changes involve adding the same logic block to multiple methods in RMSNorm and GemmaRMSNorm. While this fixes the issue, it introduces significant code duplication. My review includes a suggestion to refactor this duplicated logic to improve code maintainability.

Comment on lines +143 to +144
if post_residual_addition is not None:
residual = residual + post_residual_addition
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

This logic for handling post_residual_addition is now duplicated in at least 10 different methods across RMSNorm and GemmaRMSNorm. This introduces a significant maintainability issue.

To avoid this, I suggest centralizing the logic.

For GemmaRMSNorm, since all its forward_* methods now use the same logic, you could override the main forward method to handle post_residual_addition once. The specific forward_* methods would then not need to handle it. This would look something like:

# In GemmaRMSNorm
def forward(self, x: torch.Tensor, residual: Optional[torch.Tensor] = None, post_residual_addition: Optional[torch.Tensor] = None) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
    if residual is not None and post_residual_addition is not None:
        residual = residual + post_residual_addition
    
    # Assuming self._forward_method is the platform-specific implementation
    # Pass post_residual_addition=None as it has been handled.
    return self._forward_method(x, residual, post_residual_addition=None)

After this, you can remove the duplicated blocks you've added to GemmaRMSNorm's methods.

For RMSNorm, it's more complex as forward_native has different logic. However, the other forward_* methods in RMSNorm could be refactored to call a common private helper function to reduce duplication.

Consolidating this logic will make the code cleaner and easier to maintain.

@JustinTong0323
Copy link
Copy Markdown
Collaborator

/tag-and-rerun-ci

@JustinTong0323
Copy link
Copy Markdown
Collaborator

/tag-and-rerun-ci

1 similar comment
@JustinTong0323
Copy link
Copy Markdown
Collaborator

/tag-and-rerun-ci

@Kangyan-Zhou Kangyan-Zhou merged commit ad05782 into sgl-project:main Jan 23, 2026
264 of 283 checks passed
Johnsonms pushed a commit to Johnsonms/sglang that referenced this pull request Feb 14, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants