Install#

Prebuilt packages are available on both PyPI and Conda. For most users, we recommend installing from PyPI. For development, we recommend installing from source with Pixi.

Install via Conda (alternative)#

conda install numbast

Validate the installation (optional)#

pixi run -e test-cu12 test

Replace test-cu12 with test-cu13 if you are testing against CUDA 13.

Note

If you see errors like “cannot find header ‘cuda.h’”, please refer to FAQ for more details.

Building Documentation#

Dependencies#

Use Pixi to ensure consistent versions. In pixi.toml, the docs feature includes:

  • sphinx

  • sphinx-copybutton

  • nvidia-sphinx-theme

You also need Python and a working environment to import numbast if you want versioned builds to reflect the installed package version (optional). You can enter one with pixi shell -e test-cu13.

Build Steps#

  • Recommended: build all versions listed in docs/nv-versions.json via the Pixi task from the repository root:

    pixi run -e test-cu13 build-docs
    
  • Build only the latest version:

    pixi run -e test-cu13 build-docs latest-only
    

    Replace test-cu13 with test-cu12 if needed.

  • Optional fallback (if you are already in a prepared shell environment):

    cd docs
    ./build_docs.sh
    # or
    ./build_docs.sh latest-only
    

Artifacts are generated under:

  • docs/build/html/latest

  • docs/build/html/<version> where <version> comes from SPHINX_NUMBAST_VER or the installed numbast package version.

Notes#

  • The build script sets SPHINX_NUMBAST_VER from the installed numbast package version unless SPHINX_NUMBAST_VER is provided explicitly.

  • Output also copies versions.json and nv-versions.json, and creates a redirect index.html.