Upstream recommends a fresh conda env, but repo.anaconda.com is not reachable from Ra, so the build uses uv instead: it fetches a standalone Python 3.12 into $PREFIX/python (upstream supports >=3.9,<3.13), creates $PREFIX/venv, and runs uv pip install careless[cuda]==<version> into it. The [cuda] extra pulls the NVIDIA libraries as pip wheels, so the same install works on GPU and CPU-only Ra nodes.

uv must be on PATH on the build host (e.g. pip install --user uv).

Loading the module puts the venv's bin/ on PATH (the careless command is directly available, no activation needed) and sets CARELESS_ENV to the venv path.

Adding a new version

Add the version under versions: in files/config.yaml, then on a Ra login node: modbuild build <version> (modbuild/2.1.2).