By default cupy does not pip install the cuda toolkit (unless the [ctk] extra is specified), so we should use the sysctk extra in that situation, and make sure that we test it in our CI. Additionally, our sysctk extra currently pulls in the ctk through numba-cuda, so we need to adjust the definition of the extra to not do that, by simply removing the [cu12]/[cu13] extra
By default cupy does not pip install the cuda toolkit (unless the
[ctk]extra is specified), so we should use thesysctkextra in that situation, and make sure that we test it in our CI. Additionally, oursysctkextra currently pulls in the ctk through numba-cuda, so we need to adjust the definition of the extra to not do that, by simply removing the[cu12]/[cu13]extra