You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As discussed in #7597 (comment), the technique used in the cupy-wheel stopped working with the recent Pip release. Although we published a hotfix with a workaround, it is not behaving the same as before: users will have to remove the pip wheel cache manually in several cases. As a result, we are considering removing the package in light of the following circumstances:
setup.py is not preferable in the entire ecosystem. Running code on installation is slow and sometimes might be insecure. (Use pyproject.toml #6986 (comment))
Dependencies are often expected to be resolved statically, especially when in conjunction with Poetry or pipenv.
The variation of wheels has decreased. When introducing the cupy-wheel package, we have separate packages for each CUDA version (cupy-cuda102, cupy-cuda112, etc.). Now that we adopted the CUDA Minor Version Compatibility (Utilize CUDA Enhanced Compatibility #6730), the variation will decrease to 2 for CUDA (cupy-cuda11x and cupy-cuda12x) after dropping support for CUDA 11.1 or earlier ([RFC] Drop support for CUDA 10.2/11.0/11.1 in CuPy v13 #7557).
As discussed in #7597 (comment), the technique used in the
cupy-wheelstopped working with the recent Pip release. Although we published a hotfix with a workaround, it is not behaving the same as before: users will have to remove the pip wheel cache manually in several cases. As a result, we are considering removing the package in light of the following circumstances:setup.pyis not preferable in the entire ecosystem. Running code on installation is slow and sometimes might be insecure. (Usepyproject.toml#6986 (comment))cupy-wheelpackage, we have separate packages for each CUDA version (cupy-cuda102,cupy-cuda112, etc.). Now that we adopted the CUDA Minor Version Compatibility (Utilize CUDA Enhanced Compatibility #6730), the variation will decrease to 2 for CUDA (cupy-cuda11xandcupy-cuda12x) after dropping support for CUDA 11.1 or earlier ([RFC] Drop support for CUDA 10.2/11.0/11.1 in CuPy v13 #7557).