Installation
bolero uses pixi to manage its full environment — Python,
PyTorch + CUDA, and all dependencies are pinned in the repo, so installation is a
single command. You do not need conda or a system Python first.
Requirements: Linux (linux-64) with an NVIDIA GPU (driver CUDA ≥ 12.0).
1. Install pixi
curl -fsSL https://pixi.sh/install.sh | bash
Open a new shell (or source ~/.bashrc) so pixi is on your PATH, then check:
pixi --version
2. Clone bolero
git clone https://github.com/liuhlab/bolero.git
cd bolero
3. Install
From the repo root:
pixi install # default runtime environment (full GPU stack)
This resolves and installs every pinned dependency — PyTorch + CUDA, Ray,
flash-attn, scvi-tools — plus bolero itself. Two companion packages are pulled
straight from git during this step, so you need network/git access:
bolerodata— the dataset / model-zoo registry that maps short keys to on-disk artifacts in the lab data lake.SEACells— a lightly-repinned fork used for metacell construction (CPU).
The project defines three environments; install whichever you need:
| Environment | Install | Contents |
|---|---|---|
default |
pixi install |
runtime: full GPU stack (torch + CUDA, Ray, flash-attn) |
dev |
pixi install -e dev |
adds tests, linting, notebooks, JupyterLab |
docs |
pixi install -e docs |
lightweight, CUDA-free — builds the docs site only |
4. Run bolero
There is no activate step. Prefix commands with pixi run:
pixi run python -c "import bolero"
...or open a shell with the environment ready:
pixi shell
python -c "import bolero"
Use bolero in Jupyter
Register the environment as a Jupyter kernel named bolero:
pixi run install-kernel
Then start JupyterLab and select the bolero kernel:
pixi run jupyter lab
5. Verify
bolero.print_environments() reports the key versions and GPU status:
pixi run python -c "import bolero; bolero.print_environments()"
Expected output (versions and GPU will differ):
----- bolero environment -----
bolero : 2026.7.10
python : 3.11.15
platform : Linux-5.15.0-177-generic-x86_64-with-glibc2.35
torch : 2.4.1
torch CUDA : 12.0
CUDA available : True
GPU 0 : NVIDIA H100 80GB HBM3 (79 GB)
flash-attn : 2.6.3 (import OK)
ray : 2.34.0
numpy : 2.4.6
pandas : 2.3.3
scvi-tools : 1.4.2
transformers : 5.13.0
If CUDA available is True and flash-attn reports import OK, you are ready to go.