pfnstudio 0.8.1

CLI for the PFN Studio framework — scaffold, validate, lint, and run prior-fitted foundation model projects.

priorstudio — CLI

The command-line interface for PFN Studio, the toolkit for training prior-fitted foundation models.

Install

pip install pfnstudio

For training (requires PyTorch):

pip install "pfnstudio[torch]"

Commands

pfnstudio init <dir>            # scaffold a new FM project
pfnstudio validate <path>       # check artifacts against JSON Schema
pfnstudio lint <project>        # cross-reference + style checks
pfnstudio sample <prior.yaml>   # draw N tasks from a prior
pfnstudio run <run.yaml>        # execute a training run end-to-end
pfnstudio predict <run-dir>     # inference against a trained checkpoint
pfnstudio export <project>     # tar-gzipped project archive

Run priorstudio --help for the full list and <cmd> --help for each subcommand's flags.

What this CLI is for

PFN Studio organises every PFN project around five first-class artifacts: priors (synthetic data generators), models (block compositions), evals (benchmarks + metrics), runs (training manifests), and initiatives (research workstreams). This CLI operates on the file layout those artifacts produce — scaffolding new projects, validating them, running training, and exporting them for sharing.

The full story (concepts, architecture, examples, marketplace catalog) lives at the main repo: github.com/profitopsai/pfnstudio

License

Apache-2.0.