Moonshot open-sources MoonEP for balanced MoE expert parallelism
SeunghyunSEO7 · x · 2026-07-28
MoonshotAI has open-sourced MoonEP, an expert-parallelism communication library for keeping token loads balanced across ranks.
- It uses dynamic redundant experts to make every rank receive exactly S × K tokens, regardless of routing skew.
- The library plans redundant experts online from the current router outputs and prefetches them before expert computation.
- A GPU planning kernel handles this with negligible overhead.
- It emphasizes zero-copy and static shapes, aiming at efficient MoE execution and lower activation-memory pressure.
More from Infra
- Dedicated Hardware for Running AI Agents at Scale Arrives — cyrilzakka · 2026-09-23
- Ternary Bonsai 2 27B: 5.9GB weights retain ~95% of full-precision reasoning — cephaloform · 2026-09-23
- Qwen 27B runs 24hr unattended on one RTX5090, builds full Postgres-SpringBoot-React spreadsheet app — anglepoiselife · 2026-09-23
- OpenRoboto Shift launches: decentralized egocentric video data network for robot brains — markjeffrey · 2026-09-23
- Engineer describes designing digital circuits that recycle most of their energy — MikePFrank · 2026-09-23
- Cloudflare CTO Dane Knecht makes TIME's 2026 executives list as AI crawlers hit 52% of traffic — dinasaur_404 · 2026-09-23