Installation¶

Using pip¶

Install the core library:

pip install scribe

Install SCRIBE from source:

pip install git+https://github.com/mrazomej/scribe.git

Install CLI/Hydra tooling:

pip install "scribe[hydra]"

Using uv (recommended)¶

uv is the recommended package manager for SCRIBE development:

git clone https://github.com/mrazomej/scribe.git
cd scribe
uv sync

To include CLI/Hydra extras in the active environment:

uv sync --extra hydra

Development Installation¶

For contributors and developers:

git clone https://github.com/mrazomej/scribe.git
cd scribe
uv sync --group dev

To include CLI/Hydra tooling during development:

uv sync --group dev --extra hydra

To also install documentation dependencies:

uv sync --group dev --group docs

To install docs + CLI/Hydra tooling together:

uv sync --group dev --group docs --extra hydra

Paper analysis extras¶

The paper dependency group adds optional tooling for paper figures and benchmarks. It includes pertpy and GPU-accelerated rapids-singlecell via the CUDA 12 wheel stack, aligned with the project's jax[cuda12] install (NVIDIA driver 525+, CUDA 12.x).

uv sync --group paper

This pulls rapids-singlecell-cu12[rapids] and RAPIDS dependencies (cupy, cudf, cuml, …) from PyPI and the NVIDIA Python index. The download is large (several GB).

On GPU nodes, configure JAX before using either stack:

source scripts/setup_cuda_env.sh --gpu
python -c "import jax, rapids_singlecell as rsc; print(jax.devices()); print(rsc.__version__)"

Breaking Change Note (Hydra Boundary)¶

Base scribe installs no longer require hydra-core or omegaconf.
scribe-infer and scribe-visualize now require scribe[hydra].
If a CLI command reports missing optional dependencies, install: pip install "scribe[hydra]" (or uv sync --extra hydra).

GPU Support¶

SCRIBE uses JAX for GPU-accelerated computation. The default installation includes CUDA 12 support via jax[cuda12] (NVIDIA driver 525+). If you are on a system without a GPU, JAX will automatically fall back to CPU execution.

After uv sync, source scripts/setup_cuda_env.sh on GPU nodes to configure the environment and verify JAX can see your device.

For specific JAX installation instructions (e.g., CPU-only installs or TPU support), see the JAX installation guide.

Docker¶

A Docker environment is also available for development:

docker build -t scribe-dev .
docker run --gpus all -it scribe-dev

Verifying Installation¶

After installation, verify that SCRIBE is working:

import scribe
print(scribe.__doc__)

If JAX GPU support is available:

import jax
print(jax.devices())  # Should show GPU device(s)