Skip to content

Document expert parallelism flag differences in Claude configs - #864

Merged
functionstackx merged 1 commit into
mainfrom
claude/expert-parallel-docs
Mar 5, 2026
Merged

functionstackx merged 1 commit into
mainfrom
claude/expert-parallel-docs

Conversation

@Klaud-Cold

Copy link
Copy Markdown
Collaborator

Summary

  • Document the difference between vLLM's --enable-expert-parallel (boolean flag) and SGLang's --expert-parallel-size N (explicit integer) in both claude.yml and claude-pr-review.yml
  • vLLM/ATOM scripts should conditionally enable --enable-expert-parallel based on EP_SIZE env var, not hardcode it
  • Add review validation rule to flag hardcoded --enable-expert-parallel in PR reviews

Context

vLLM serve does not accept --expert-parallel-size directly — it uses the boolean --enable-expert-parallel flag and determines EP size automatically. Scripts should use the conditional EP_SIZE check pattern rather than always enabling expert parallelism.

Referenced from PR #757 review discussion.

Generated with Claude Code

Document the differences between vLLM's --enable-expert-parallel (boolean flag)
and SGLang's --expert-parallel-size N (explicit integer) in both claude.yml and
claude-pr-review.yml. Scripts should conditionally enable --enable-expert-parallel
based on the EP_SIZE env var rather than hardcoding it.

Co-authored-by: functionstackx <functionstackx@users.noreply.github.com>
@Klaud-Cold
Klaud-Cold requested a review from a team March 5, 2026 04:06
@functionstackx
functionstackx merged commit 05fc951 into main Mar 5, 2026
@functionstackx
functionstackx deleted the claude/expert-parallel-docs branch March 5, 2026 04:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

2 participants