Running Experiments

The typical workflow is two steps: configure a run → run (or submit) it.

theseus configure <job> config.yaml             # make config

theseus run <name> config.yaml <output_dir>     # run locally
theseus submit <name> config.yaml               # or send to remote infra

Before you get started, you can run:

theseus jobs

This prints a table of every registered experiment. Pick one — for the examples below we'll use gpt/pretrain.

Choose a workflow:

• Local — run on your current machine
• Remote (SSH & SLURM) — submit to a plain SSH host or SLURM cluster
• Remote (Volcano / K8s) — submit to a Kubernetes cluster with Volcano
• Interactive REPL — launch a remote Jupyter session