Agent skill · NVIDIA
nemo-mbridge-perf-tp-dp-comm-overlap
Operational guide for enabling TP, DP, and PP communication overlap in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.
What it needs
About 1k tokens when loaded.
What this skill does
TP / DP / PP Communication Overlap Skill For stable background and recommendation level, see: @docs/training/communication-overlap.md Enablement Minimal Bridge override: Optional TP preset: Precision knobs belong to mixed precision: Code Anchors Bridge overlap gating: PP overlap selection: DP overlap defaults: Launch-time env tuning: Pitfalls 1. TP overlap silently disables itself if sequenceparallel=False or Transformer Engine is unavailable. 2. PP overlap is not enabled for all PP cases. Bridge only auto-selects overlapp2pcomm=True when PP > 1 and VPP > 1. 3. bucketsize is a parameter-count knob, not a byte-size knob. 4. gradreduceinfp32 and fp8paramgather should be set through mixed precision, not as standalone DDP tuning first. 5. CUDADEVICEMAXCONNECTIONS and LayerNorm SM margin are launch-time plugin settings, not CommOverlapConfig fields. Verification Use the checked-in overlap unit coverage first: Optional second check if nemorun is available: Success criteria: first command reports 26 passed second command validates plugin-owned env wiring when not skipped
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills NVIDIA/nemo-mbridge-perf-tp-dp-comm-overlap