
Guides tuning of MoE expert-parallel communication overlap in Megatron Bridge
- Open source
- 4K GitHub stars
- AI models and LLM tooling


Guides tuning of MoE expert-parallel communication overlap in Megatron Bridge

Guides enabling and verifying TP, DP and PP communication overlap in Megatron-Bridge

Guides an agent through evidence-gated MoE performance tuning in Megatron Bridge

Guides choice of MoE token dispatcher (alltoall, DeepEP, HybridEP) by hardware and EP size

Guides validating and configuring CUDA graph capture in Megatron Bridge training runs

Guides an agent through reducing peak GPU memory in Megatron Bridge training runs

Configures fault tolerance, straggler detection and restarts for Megatron Bridge training

Guides an agent through multi-node Slurm sbatch jobs and debugging their failures

Guides configuring and measuring NeMo Relay adaptive plugin telemetry, hints

Guides integrating PhysicsNeMo ShardTensor domain parallelism into multi-GPU training

Guides choosing and configuring NeMo Relay 0.6/0.7 observability exporters and subscribers

Practitioner runbook for training and serving large models, from GPUs to NaN loss spikes
The closest alternatives to Nemo mbridge perf expert parallel overlap are Nemo mbridge perf moe comm overlap, Nemo mbridge perf tp dp comm overlap, Nemo mbridge perf moe optimization workflow, Nemo mbridge perf moe dispatcher selection and Nemo mbridge perf cuda graphs.
Yes. Nemo mbridge perf moe comm overlap, Nemo mbridge perf tp dp comm overlap, Nemo mbridge perf moe optimization workflow have a free or open-source option.