Contributing to versions to pypi

[pfmephisto]

Hi,
I am running in to issues when trying to install torch-points-kernels using pip, with the issue being that no compatible build can be found there.

I have since compiled my own build from sources but was wondering if there is a way to contribute that to pypi so that future installs might be simpler.

[clee-ai]

Hi! Thanks for your interest. Unfortunately, due to the various torch/cuda combinations, we can't host wheels directly on pypi. What other projects do is self-host the wheels with all such combinations (ex https://data.pyg.org/whl/) but currently we are not set up to do this. Currently, @Stakhan is working to get conda installation working, and in the future I'd be interested to get wheels working as well, but there's infrastructure we'd have to set up around that.

[pfmephisto]

@CCInc: Thanks for getting back to me on this one.

@Stakhan: Am I able to help you with anything regarding getting Conda working?
It's not that I know a whole lot about the process but, I'd be willing to put some time and effort into it.
At least with the selfish reason that I then can get rind of my custom scripts/notes on the installation 😀

[clee-ai]

@pfmephisto I had a chance to look at this. If you're able to test out my conda build with a few different permutations of python/pytorch versions, that'd be a huge help! Mostly concerned about the CUDA working correctly.

I'll close this issue for now, and you see how to install and follow up in this issue: #89 (comment)