Prepare a source checkout

Maturity: implemented

RFFM is not distributed as a Python package. Work from a source checkout and use the repository environment that Training CI exercises.

Requirements

  • Git

  • uv

  • Python 3.11 or later; CI installs the version selected by the repository

From the repository root:

uv python install 3.11
uv sync --project analysis/eda --group dev

The analysis/eda project currently supplies the shared Python dependencies, including PyTorch, Hydra, Pydantic, HDF5, Parquet, YAML, pytest, and Ruff. This is a repository development environment, not a promise about a future package’s dependency surface.

Verify imports and the command surface

PYTHONPATH=src:. uv run --project analysis/eda \
  python -m rffm.cli.main pretrain --help

The command must show the required --experiment, --run-id, and --attempt-id options. It does not load data or start training when invoked with --help.

Run the bounded training test suite with the same environment:

PYTHONPATH=src:. uv run --project analysis/eda pytest -q training/tests

The tests use small temporary canonical bundles and CPU execution. Passing them does not prove that the complete RadioML2018 dataset or Google Cloud credentials are available on the current machine.

See Also