Leoda

Home

❯

Big Feature

❯

DModel Training optimization

DModel Training optimization

Aug 12, 20261 min read

DModel project

  • DModel 简单摸底 0. goin ✅ 2026-06-30
  • 梳理当前Megatron内的 ep overlap 1. MoE EP Overlap ✅ 2026-06-30
  • 输出详细的DModel backward Wgrad 计算与 dispatch backward的通信的 overlap 1. backward ep overlap design(PASS ✅ 2026-07-15
  • 输出一份拆分 ep 的方案 2. DModel TBO

Graph View

Created with Quartz v4.5.2 © 2026

  • GitHub
  • Discord Community