it is shown that the latest version of the nvidia-ml-py package is 10.418.84.
However, when I run:
pip install nvidia-ml-py
I get a very old version (375.53) installed:
Successfully installed nvidia-ml-py-375.53.1
Clearly, this should not be the case and was a source of much confusion, especially as the older version doesn’t support Python 3!
The core problem is here: https://pypi.org/simple/nvidia-ml-py/
Pip sees the inconsistently-named version “375.53” and decides that it’s much newer than “10.418.84” as 375 > 10. Can the link to the nvidia-ml-py-375.53.tar.gz version please be removed so that pip works correctly again? Or can the version number become for example 7.375.53?
I am aware that I can just do:
pip install nvidia-ml-py==10.418.84
But this means that every time anyone installs nvidia-ml-py they first have to understand that this version problem even exists, look up the versions on the website, and manually copy out the version number. When newer versions are released it also restricts easy updates due to the fixed version number.
Dear NVIDIA pypi package maintainers, please remove nvidia-ml-py-375.53 from pypi so that pip install nvidia-ml-py will install the correct latest version.
@mnicely I wrote directly to nvml-bindings@nvidia.com, but I got access denied because this seems to be an internal mailing list. Can you please forward it to whom it may concern again?
Hello!
I’m one of the maintainers for the package on PyPi.
Thanks for bringing this to my attention, I am working on fixing that versioning issue.
PyPi is being a tad annoying about it, since it’s not listed on the release page so it’ll take me a moment to get it resolved.