Installation
Requirements
Python >= 3.11, < 3.14
PyTorch >= 2.8.0 (with CUDA support recommended)
A CUDA-capable GPU (for the Triton/CUDA hypergeometric kernels)
Install from PyPI
pip install pgcuts
Install from source
git clone https://github.com/ayghri/pgcuts.git
cd pgcuts
pip install .
For development:
pip install -e ".[experiments]"
Verify installation
from pgcuts import HyCut
print("PGCuts installed successfully!")
# Quick GPU check
import torch
from pgcuts import Hyp2F1
z = torch.rand(10, device="cuda")
out = Hyp2F1.apply(-10, 1.0, 2.0, z)
print(f"2F1 output: {out}")