janus-moe-disaggregation

【效率与协作】Enable scalable MoE inference by disaggregating attention and expert layers onto independent GPU sub-clusters. Use adaptive two-phase communication, activation load-balanced scheduling, and activation-aware expert management. Achieve 3.9× higher per-GPU throughput than state-of-the-art systems.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准