i have some questions about nvshmem4py
- will nvshmem4py be released separately, or will it be released with nvshmem. I often wait for a long time for nvshmem upgrade, but maybe nvshmem4py can be released in a faster pace until it’s stable?
- about the interface: will nvshmem4py cover all interfaces from nvshmem C API, or just as it is now? Some interfaces are not covered, such as
nvshmem_team_split_strided and nvshmem_team_split_2d
- is this wheel maintained by official team: piwheels - nvidia-nvshmem-cu12. here is a 0.1.1, but here( Index of /compute/nvshmem/redist/nvshmem_python/source) there is only 0.1.0
Thank you for your interest in the NVSHMEM4Py project:
- will nvshmem4py be released separately, or will it be released with nvshmem. I often wait for a long time for nvshmem upgrade, but maybe nvshmem4py can be released in a faster pace until it’s stable?
NVSHMEM4Py is released as part of NVSHMEM releases today, but with a separate versioning policy and a somewhat self-contained codebase.
- about the interface: will nvshmem4py cover all interfaces from nvshmem C API, or just as it is now? Some interfaces are not covered, such as
nvshmem_team_split_strided and nvshmem_team_split_2d
Our goal is to provide additional features in the future. We are trying to prioritize important use cases and we will do our best to help our users achieve their goals as easily as possible. Your feedback here on which sets of APIs to prioritize is valuable to us.
We only maintain the wheels hosted on PyPI (e.g. nvshmem4py-cu12 · PyPI) and the NVIDIA channel of Conda-Forge.