slurptuna

Run Optuna hyperparameter optimization on Slurm (HPC clusters) without boilerplate.

slurptuna is a simple way to run Optuna on Slurm clusters without writing sbatch scripts or managing distributed workers.

Running Optuna in distributed mode on a Slurm cluster (HPC) normally requires writing custom job scripts, managing shared storage, and coordinating distributed workers manually. slurptuna wraps this behind a single function call.

How slurptuna runs Optuna on Slurm

You write a loss function decorated with @loss, then call optimize_run. slurptuna handles:

• Submitting per-trial chunk array jobs via sbatch
• Collecting and aggregating seed-level results
• Storing Optuna trial state in a local SQLite database
• Writing a summary.json with the best result when done

Install

pip install slurptuna

Or with uv:

uv add slurptuna

Next steps

• Quickstart — write your first loss and run it
• Distributed Mode — how chunk/reduce jobs work, what knobs to turn, and benchmark numbers
• Participant-wise Fitting — fitting one parameter set per participant

Full docs: younesstrittmatter.github.io/slurptuna