Launch and troubleshoot multi-node training jobs.
What it does
Launches distributed Megatron-LM training on a SLURM cluster with submission scripts, environment configuration, monitoring, and failure diagnosis.
When to use it
Use when submitting SLURM jobs, writing or debugging sbatch scripts, configuring multi-node training, or diagnosing job failures.
How to use it
Give it a Megatron-LM worktree to configure; it prepares a SLURM submission script in that worktree.
What you provide
- An existing project
Uses
SL
Latest review
What you need · 3
Requires a cluster login with permission to submit jobs to a GPU partition.
Requires uv installed and a synchronized worktree environment visible to every node before submission.
Access · 2
This skill
SLURM
Execute
Starts distributed training jobs
Write
Submits jobs to SLURM
uv
Execute
Installs training environment dependencies