Installation

PrexSyn requires Python 3.11 or newer. The published PrexSyn Engine wheels currently target Linux x86-64; CUDA is recommended for model inference and required by the training script.

Install for the examples

git clone https://github.com/luost26/prexsyn.git
cd prexsyn
uv sync

uv creates the environment from pyproject.toml and uv.lock. Verify both packages:

uv run python -c "import prexsyn, prexsyn_engine; print(prexsyn.__version__)"

Run commands from the repository root. Shipped configuration files use paths relative to that directory.

Install in another project

python -m pip install "prexsyn @ git+https://github.com/luost26/prexsyn.git"

PrexSyn Engine is installed as a dependency. A PrexSyn release on PyPI is not yet available.

Choose a device

Example scripts default to --device cuda. Use --device cpu when CUDA is unavailable, but expect substantially slower inference. To select a different PyTorch/CUDA build, follow the uv PyTorch guide before syncing the environment.

Optional pathway images

Install the Graphviz system package to use --draw-output-dir or item.get_image(). Python dependencies alone are not enough because pydot calls the Graphviz executable.