MoonshotAI open-sources MoonEP for perfectly balanced expert parallelism
teortaxesTex · x · 2026-07-27
MoonshotAI open-sourced MoonEP, an expert-parallelism communication library designed to keep token loads perfectly balanced across ranks.
- The repo says it uses dynamic redundant experts to ensure each rank receives exactly S × K tokens, even when routing is highly skewed.
- It plans redundant experts online from current router outputs, prefetches them before expert computation, and reduces gradients back to home ranks in the backward pass.
- The implementation emphasizes near-optimal GPU planning, zero-copy execution, and static shapes.
More from Infra
- fmgo: call Apple's on-device Foundation Models from Go with no CGO and no Swift — Super_Run_8466 · 2026-09-23
- Huawei unveils Peerium architecture: nested BSP unifies million processors into one computer — Dr_Singularity · 2026-09-23
- Grok explains why DeepSeek picked DualPipe + ZeRO-1 over ZeRO-3 on 2048 H800s — TheZachMueller · 2026-09-23
- AI costs fall 47% per quarter, 4x faster than DNA sequencing: Epoch AI — daveholtz · 2026-09-23
- M5 Ultra LLM test: 4x faster prompt processing, but double the power draw — DigitalguyCH · 2026-09-23
- $500 of Dell OptiPlexes become a diskless netboot lab where AI agents can't brick the hardware — colinmcnamara · 2026-09-23