
Mcore run on slurm
SkillLaunch distributed Megatron-LM training jobs on SLURM clusters with sbatch scripts
- Open source
- 4K GitHub stars
- Developer tools


Launch distributed Megatron-LM training jobs on SLURM clusters with sbatch scripts

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

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

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

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

Runs TAO GPU jobs on a remote SLURM cluster over SSH with sbatch and Pyxis containers

Runs iterative NeMo-RL experiments with git branches and TSV logs as a research ledger

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

Guides validating and enabling MoE expert-parallel communication overlap

Guides an agent through launching and debugging NeMo-RL training runs on Kubernetes

Guides configuring and measuring NeMo Relay adaptive plugin telemetry, hints

Practitioner runbook for training and serving large models, from GPUs to NaN loss spikes
The closest alternatives to Nemo mbridge multi node slurm are Mcore run on slurm, Nemo mbridge perf moe optimization workflow, Nemo mbridge resiliency, Nemo mbridge perf memory tuning and Nemo mbridge perf cuda graphs.
Yes. Mcore run on slurm, Nemo mbridge perf moe optimization workflow, Nemo mbridge resiliency have a free or open-source option.