Skip to content

Support Waterfill with MegaMoE backend#3

Closed
xutizhou wants to merge 2 commits into
fix/megamoe-shared-fusion-quantfrom
fix/megamoe-waterfill
Closed

Support Waterfill with MegaMoE backend#3
xutizhou wants to merge 2 commits into
fix/megamoe-shared-fusion-quantfrom
fix/megamoe-waterfill

Conversation

@xutizhou

@xutizhou xutizhou commented Jun 5, 2026

Copy link
Copy Markdown
Owner

Summary

  • allow Waterfill to preserve an explicit MegaMoE backend instead of forcing DeepEP
  • allow the MegaMoE env path to be selected before Waterfill fallback logic runs
  • use a rank-local shared expert slot semantic helper for TopK, FusedMoE, and DeepSeek shared expert fusion call sites
  • add server-args tests for explicit MegaMoE and MegaMoE env selection under Waterfill

Dependency

Depends on sgl-project#27349, which adds the shared expert fusion loading compatibility needed by the current DeepSeek-V4-Flash FP4 test path.

Testing

  • python3 -m compileall -q python/sglang/srt/layers/moe/fused_moe_triton/layer.py python/sglang/srt/models/deepseek_v2.py python/sglang/srt/layers/moe/topk.py python/sglang/srt/layers/moe/utils.py python/sglang/srt/server_args.py
  • B200 MMLU subset: pure MegaMoE accuracy 0.8509861213; Waterfill+MegaMoE accuracy 0.8531775018

@xutizhou
xutizhou force-pushed the fix/megamoe-shared-fusion-quant branch from 7d4f3c3 to c68edc2 Compare June 5, 2026 06:55
@xutizhou
xutizhou force-pushed the fix/megamoe-waterfill branch from 40c2b3e to f57c56d Compare June 5, 2026 06:55
@xutizhou

xutizhou commented Jun 5, 2026

Copy link
Copy Markdown
Owner Author

Closing this fork-internal stacked PR. Reopening as an upstream PR from xutizhou:fix/megamoe-waterfill to sgl-project:main, depending on sgl-project#27349.

@xutizhou xutizhou closed this Jun 5, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant