Nemo mbridge perf cuda graphs vs Nemo mbridge perf moe optimization workflow
Nemo mbridge perf cuda graphs is an open-source agent skill for developer tools. Nemo mbridge perf moe optimization workflow is an open-source agent skill for developer tools.
- Nemo mbridge perf cuda graphs is best for enable cuda graphs for a dense model training run in megatron bridge.
- Nemo mbridge perf moe optimization workflow is best for diagnose a throughput regression after a config or commit change in a moe training run.
- What it is
- Nemo mbridge perf cuda graphs is an open-source agent skill for developer tools.Nemo mbridge perf moe optimization workflow is an open-source agent skill for developer tools.
- Best for
- Enable CUDA graphs for a dense model training run in Megatron BridgeDiagnose a throughput regression after a config or commit change in a MoE training run
- Kind
- SkillSkill
- Pricing
- Open sourceOpen source
- Platforms
- Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf, GooseClaude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf, Goose
- GitHub stars
- 4K4K
- Last commit
- Nov 2026Nov 2026
- License
- Apache-2.0Apache-2.0
- Open source
- Good for
- Enable CUDA graphs for a dense model training run in Megatron Bridge; Pick a graph scope for a dropless MoE pretraining configuration; Diagnose a crash caused by a CUDA graph config changeDiagnose a throughput regression after a config or commit change in a MoE training run; Plan a full MoE throughput tuning sweep for a large expert-parallel model; Choose between alltoall, DeepEP and HybridEP dispatchers for a cluster
What Nemo mbridge perf cuda graphs does
Two graph implementations
Explains the local full-iteration wrapper and the Transformer Engine per-module scoped graphs, and when each applies.
Scope selection guidance
Recommends starting scopes for dense, dropless MoE and VLM models and how graphs interact with selective recompute.
Configuration constraints
Lists required settings and conflicts that the framework enforces, such as RNG tracker and NaN-check requirements.
Pitfalls and memory notes
Covers memory overhead, variable-length sequence limits, cleanup of graph objects and workload-dependent speedups.
Bring-up and validation steps
Gives an ordered procedure for enabling graphs and comparing replay timing against eager runs, plus test commands.
What Nemo mbridge perf moe optimization workflow does
Measurement contract setup
Lists the versions, topology, batch shape and metric windows that must stay fixed for a valid comparison.
Three Walls diagnosis
Classifies throughput limits as memory, communication, host overhead or compute and maps each to fixes.
Parallel Folding guidance
Explains separate attention and MoE parallel meshes and when to prefer expert parallelism over extra tensor parallelism.
Dispatcher and overlap tuning
Compares alltoall, DeepEP and HybridEP dispatchers and the communication overlap options to test.
Lower-precision candidate matrix
Maps hardware generations to FP8, MXFP8 or NVFP4 candidates that follow a verified BF16 baseline.
Profiling and validation steps
Describes interval-union overlap analysis and the evidence bundle required before accepting a winner.
Nemo mbridge perf cuda graphs vs Nemo mbridge perf moe optimization workflow: questions
Is Nemo mbridge perf cuda graphs or Nemo mbridge perf moe optimization workflow cheaper?
Both are open source. Compare current plans on each official site.
What is the difference between Nemo mbridge perf cuda graphs and Nemo mbridge perf moe optimization workflow?
Nemo mbridge perf cuda graphs is an open-source agent skill for developer tools. Nemo mbridge perf moe optimization workflow is an open-source agent skill for developer tools.
