GPU arrays (via CuPy) when possible, otherwise CPU arrays (via NumPy)
Code: https://github.com/static-frame/hyray
Packages: https://pypi.org/project/hyray
The hyray library provides a single interface to use both NumPy and CuPy, depending on the availability of CuPy, CUDA, or sufficient memory. With hyray a single implementation can run with CuPy when CuPy is available (on a local machine with a GPU), and still work with NumPy when on GPU or insufficient memory is available (such as on continuous integration servers).