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
- Moonshot open-sources MoonEP as open models vs closed labs debate intensifies — KyeGomezB · 2026-07-27
- Kimi K3’s 2.5x scaling-law gain draws praise for training efficiency — andrew_n_carr · 2026-07-27
- Kimi K3 goes live on Nebius with 1M-token context and a 57 AA score — teortaxesTex · 2026-07-27
- AI agent finds a longstanding Bun Node-compat bug in `child_process.spawn` — steipete · 2026-07-27
- llama.cpp adds support for Nanbeige4.2 in pull request 25994 — pmttyji · 2026-07-27
- Open-sourced Kimi K3 speculator lifts single-stream throughput from 118 to 370 tok/s — vllm_project · 2026-07-27