Convert single-node Megatron Bridge scripts into multi-node Slurm sbatch jobs.
npx skills add https://github.com/NVIDIA/skills --skill multi-node-slurm
Please help me install this Agent Skill. Skill: multi-node-slurm Source: https://github.com/NVIDIA/skills/tree/main/skills/Megatron-Bridge/multi-node-slurm Command: npx skills add https://github.com/NVIDIA/skills --skill multi-node-slurm
Convert single-node Megatron Bridge scripts into scalable multi-node Slurm sbatch jobs and provide structured debugging guidance for common multi-node failures.
Use when scaling a Megatron-Bridge script from a single node to a multi-node Slurm job, following the two-phase srun workflow.
High-intent search queries and answers about installing and using this skill.
You can convert single-node Megatron Bridge scripts to multi-node Slurm sbatch jobs by applying the two-phase srun-native or uv-run launch pattern, which derives distributed initialization variables like RANK and MASTER_ADDR from Slurm environment variables.
Scaling distributed training across Slurm nodes is best achieved by generating a structured sbatch job that manages containerized environments, resource allocation, and log organization while ensuring proper distributed initialization for reliable multi-node operations.
Yes, the multi-node Slurm workflow supports both srun-native, which is the preferred method, and uv run torch.distributed as a legacy approach to initialize distributed training across allocated cluster nodes.
You can debug NCCL timeouts and OOM sizing for MoE models during distributed training by utilizing the built-in debugging assistance that provides structured guidance for common multi-node failures and interactive allocation workflows.
Your distributed training job may fail to initialize across multiple Slurm nodes when environment variables like WORLD_SIZE, LOCAL_RANK, and MASTER_ADDR are not properly derived and configured from the Slurm environment during the launch phase.
yo-stevenConvert single-node PyTorch distributed launches into multi-node Slurm sbatch jobs.
sayalinvidiaConvert single-node Megatron Bridge scripts into multi-node Slurm sbatch workflows.
Delay-n-daysManage a unified Python virtual environment for workspace skills using uv.
afel6Enable distributed training for PyTorch scripts with automatic device placement.
