Inclusion of nvfp4 blockscale in EPLB Rebalance

[wenscarl]

Motivation

#13715 exclude block_scale in the EPLB rebalance. This PR add it back since it's expert dependent.

Modifications

Accuracy Tests

benchmark:
type: "gpqa"
num_examples: 198
repeat: 8
num_threads: 128
max_tokens: 65536

Without EPLB:

Repeat: 8, mean: 0.785
Scores: ['0.803', '0.793', '0.783', '0.778', '0.783', '0.798', '0.768', '0.778']
====================
Writing report to /tmp/gpqa_deepseek-ai_DeepSeek-R1.html
{'chars': np.float64(24479.929292929293), 'chars:std': np.float64(18990.288974058105), 'score:std': np.float64(0.41573970964154905), 'scores': ['0.803', '0.793', '0.783', '0.778', '0.783', '0.798', '0.768', '0.778'], 'mean_score': np.float64(0.7853535353535352)}

With EPLB:

Repeat: 8, mean: 0.790
Scores: ['0.803', '0.793', '0.793', '0.798', '0.773', '0.768', '0.783', '0.808']
====================
Writing report to /tmp/gpqa_deepseek-ai_DeepSeek-R1.html
{'chars': np.float64(23318.530303030304), 'chars:std': np.float64(16233.483662940014), 'score:std': np.float64(0.3938098725175339), 'scores': ['0.803', '0.793', '0.793', '0.798', '0.773', '0.768', '0.783', '0.808'], 'mean_score': np.float64(0.7897727272727272)}

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

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]

Summary of Changes

Hello @wenscarl, 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 oversight where nvfp4 blockscale parameters were incorrectly excluded from the Mixture of Experts (MoE) load balancing process. By re-including these expert-dependent parameters, the system can achieve a more accurate and effective rebalance, ensuring that all relevant components contribute to the load distribution logic.

Highlights

• Re-inclusion of nvfp4 blockscale parameters: This pull request re-includes nvfp4 blockscale parameters in the Expert Parallel Load Balancing (EPLB) rebalance mechanism. These parameters were previously excluded but are now added back due to their expert-dependent nature.
• Modification to MoE weight parameter filtering: The filter_moe_weight_param_global_expert function in python/sglang/srt/layers/moe/utils.py has been updated to remove the condition that excluded parameters ending with _blockscale_swizzled. This change ensures these parameters are correctly identified as global expert MoE weight parameters for rebalancing.

🧠 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. ↩

[gemini-code-assist]

Code Review

The pull request correctly re-includes _blockscale_swizzled parameters in the filter_moe_weight_param_global_expert function, aligning with the motivation that these parameters are now considered expert-dependent. This change ensures that the filtering logic correctly identifies parameters that do not require global experts, which is consistent with the stated purpose of the function and the PR's goal to add back block_scale for EPLB rebalance.

[hlu1]

Can you add accuracy test results?

[Fridge003]

/tag-and-rerun-ci