Nash Brown(@nash_c_brown):We’ve open-sourced the MoE megakernel we use to train models on NVL72s. A few of my favorite details: bf16 and mxfp8 support, pull-based dispatch, configurable overlap granularity, full determinism, and no CPU-GPU syncs. Read more in our blog and code!

Nash Brown

@nash_c_brown

cs @Stanford @HazyResearch, ml @cursor_ai

参加 August 2025

148 フォロー中 1.1K ファン

Nash Brown@nash_c_brown

2026.08.04 16:20

We’ve open-sourced the MoE megakernel we use to train models on NVL72s. A few of my favorite details: bf16 and mxfp8 support, pull-based dispatch, configurable overlap granularity, full determinism, and no CPU-GPU syncs. Read more in our blog and code!

285

コミュニティへ転送

人気のあるユーザー