Installation¶

BayesMBAR is written in pure Python and is published on the Python Package Index (PyPI), so you can install it with virtually any Python package manager. All of its dependencies, including JAX, are installed automatically. By default this gives you the CPU build of JAX; if you have a GPU available, we strongly recommend using a GPU build, as it significantly speeds up the calculations.

1. CPU only¶

pip install bayesmbar

# add BayesMBAR to a uv-managed project
uv add bayesmbar

# or install it into the active environment
uv pip install bayesmbar

2. NVIDIA GPU (Linux)¶

Install the cuda extra, which pulls in the CUDA 12 build of JAX:

pip install "bayesmbar[cuda]"

# add BayesMBAR to a uv-managed project
uv add "bayesmbar[cuda]"

# or install it into the active environment
uv pip install "bayesmbar[cuda]"

3. Other hardware¶

For other accelerators (e.g., TPUs, AMD GPUs, or a different CUDA version), first install the matching JAX build by following the instructions on the JAX website, then install BayesMBAR as in the CPU-only case above. The JAX build you installed will be kept.

That's it. You can verify the installation by importing the package:

python -c "from bayesmbar import BayesMBAR, FastMBAR; print('BayesMBAR is installed')"