Getting started

Three pages, in order. Start to finish is about fifteen minutes on a cluster you already have access to.

• Installation — the package, the optional extras, and what each one is for

• Quickstart — launch a swarm, run a batch job against it, tear it down

• Your first custom job — write a SwarmJob subclass and run it from both the CLI and a script

Once those work, Guides covers doing it properly: checkpointing long runs, choosing a data backend, and the platform-specific details for Slurm and Lepton.