Smoolak [1] filed an adoption request for python-xgboost-cuda [2]: I maintain tvm and would like to adopt python-xgboost-cuda as an alternative XGBoost dependency for its tests. The current 3.0.0 build fails during CMake's OpenMP_CUDA detection with CUDA 13.4 and CMake 4.4. I would start by building 3.4.1 and investigating the CUDA/OpenMP failure. I would keep CUDA support enabled in the build and run upstream CPU tests, followed by tests of training, prediction and model serialization after installation. I would document the GPU testing limits and track Arch's compiler and CUDA changes when maintaining the PKGBUILD.
[1] https://aur.archlinux.org/account/Smoolak/ [2] https://aur.archlinux.org/pkgbase/python-xgboost-cuda/
