Installation

The steps below mirror the README.

Create and activate a virtual environment (make sure you use the right python version = 3.10, 3.11 or 3.12. I highly recommend 3.12)

python -m venv venv
source venv/bin/activate

Install gridfm-graphkit from PyPI. torch-scatter wheels are not on PyPI. pip does not read [tool.uv] find-links in pyproject.toml, so pass the PyG index on the command line. The package pins torch>=2.10,<2.13, the newest line with prebuilt scatter wheels. torch==2.12.* keeps pip on that same line.

pip install gridfm-graphkit "torch==2.12.*" \
    --find-links https://data.pyg.org/whl/torch-2.12.0+cpu.html

The URL above is the CPU index. For CUDA, swap cpu for the matching tag (cu126, cu118, …). Then install torch-scatter for the torch version pip selected:

TORCH_VERSION=$(python -c "import torch; print(torch.__version__.split('+')[0])")
pip install torch-scatter -f "https://data.pyg.org/whl/torch-${TORCH_VERSION}+cpu.html"

When no prebuilt wheel exists for your platform, install torch first (still inside >=2.10,<2.13) and build scatter against it. This needs a C++ compiler:

pip install "torch>=2.10,<2.13"
pip install torch-scatter --no-build-isolation

For documentation generation and unit testing, install the dev and test extras the same way (gdown is in the test extra):

pip install "gridfm-graphkit[dev,test]" "torch==2.12.*" \
    --find-links https://data.pyg.org/whl/torch-2.12.0+cpu.html