MoE Training Optimization Workflow

Systematic workflow for MoE training optimization in Megatron Bridge, based on the Megatron-Core MoE paper. Covers the Three Walls framework, parallel folding, recompute strategy, dispatcher choice, and CUDA-graph bring-up.

Source: github.com/NVIDIA/skills/tree/HEAD/skills/nemo-mbridge-perf-moe-optimization-workflow · License: Other