[fix] logical_to_all_physical_map index 256 is out of bounds in EP parallel.

[MiterV1]

[fix] logical_to_all_physical_map index 256 is out of bounds in EP parallel.

fix bug #6625.

Motivation

Modifications

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]

Hello @MiterV1, 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!

Summary of Changes

Hello team, gemini-code-assist here to provide a summary of this pull request. This PR addresses a bug (linked as #6625) where the shared_experts_fusion optimization was being enabled by default for DeepSeek V3/R1 models under specific conditions (SM version >= 90, fp8, etc.). The change removes the code that automatically enabled this optimization, ensuring it is not turned on by default in this specific path. The intent is to fix the reported bug and potentially allow users to control this setting explicitly rather than having it enabled automatically.

Highlights

• Bug Fix: Fixes issue #6625 related to the default behavior of shared_experts_fusion.
• Default Optimization Behavior Change: Disables the automatic enabling of the shared_experts_fusion optimization for DeepSeek V3/R1 models when running on SM90 (compute capability 9.0) with fp8 and other specific conditions.
• Code Removal: Removes the specific lines of code responsible for setting self.n_share_experts_fusion and global_server_args_dict["n_share_experts_fusion"] within the determine_n_share_experts_fusion method under the affected conditions.

Changelog

• python/sglang/srt/models/deepseek_v2.py
  • Removed lines 1693 and 1694 which previously set self.n_share_experts_fusion and global_server_args_dict["n_share_experts_fusion"] to self.tp_size when specific conditions (SM version >= 90, fp8, etc.) were met.
  • Modified the log message on line 1695 (previously 1697) to indicate that the shared experts fusion optimization is not enabled in this path, reflecting the removal of the auto-enablement logic.

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.

---

A bug was found,
A default changed,
Code is cleaner now,
Optimization waits.

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 correctly addresses the issue of automatically enabling shared_experts_fusion for DeepSeek V3/R1 models under specific conditions (fp8 on SM >=90). By removing the lines that unconditionally set n_share_experts_fusion to tp_size, this change ensures that the user-configured value for n_share_experts_fusion (or its default) is respected, aligning with the goal that this setting should be user-controlled.

The modification to the log message also accurately reflects this new default behavior, stating that the optimization is not enabled by default under these conditions.

Overall, the changes are clear, concise, and directly achieve the intended bugfix. Well done!

Merge Readiness

The changes in this pull request appear to be correct and align with the stated objectives. The code is clear and effectively disables the automatic enabling of shared_experts_fusion by default, allowing user-defined configurations to take precedence. I believe this PR is in good shape to be merged. As I am an AI, I am not authorized to approve pull requests, so please ensure other reviewers take a look before merging.

[SzymonOzog]

can confirm that if fixes the issue, thanks a lot

[Alcanderian]

@BBuf Is it better to add a standalone server args to control it? Like --disable-cuda-graph.

[BBuf]

LGTM.

[zyksir]

For now, this PR just disables shared expert fusion in --enable-ep-moe mode? Is my understanding corrent?

[MiterV1]

For now, this PR just disables shared expert fusion in --enable-ep-moe mode? Is my understanding corrent?

yes,you're right.

[MiterV1]

hi @zhyncs,
Pls check my patch.
Thanks a lot.