Installation commands in PyPi page are not working for with CUDA cases

[tatban]

I have all the requirements installed along with Python 3.6 and PyTorch 1.7.1 with CUDA 11.0. Now when I run
pip3 install rational_activations it installs properly but while importing it gives a warning saying that it is not CUDA optimized version and doesn't use cuda. After a bit of search I got to see the command for CUDA version at PyPi page (https://pypi.org/project/rational-activations/). But when I run that
pip3 install rational_activations_cu110 it says :

ERROR: Could not find a version that satisfies the requirement rational_activations_cu110 (from versions: none)
ERROR: No matching distribution found for rational_activations_cu110

Infact for other CUDA versions too the whl files are not found and those existing links in PyPi page are all broken. So, humble request is to provide at least a few different versions (cuda / non-CUDA) as it is supposed to be as per the PyPi page, without just asking us to build from the source (which is anyways not so clean either due to several low level architectural dependencies).

[k4ntz]

Hi. Thanks for your feedback.
PyTorch just changed the way you can install packages that depend on theirs and I have to update accordingly.

I will find a solution to have a PyPi installable package again, but in the meantime, the best solution I can offer is to use the setup.py script.
If you know how to deal with this, please let me know, we could cooperate.

(So far, I have been using a script that compiles rational-activation in manylinux version, but then removes every PyTorch file, as most of these are included in the PyTorch library, that should be installed if you use rationals with PyTorch.
If I don't do so, I obtain a super heavy package, that I can't push to PyPi).