Adds Github Action for PyPI release!

[purepani]

Depends on #643.

This adds a github action for releasing to pypi. If you tag a commit with v*, the action will be triggered, and it will try and publish it. You'll have to set up this action as a "trusted publisher" on your pypi account for this to work.

This uses an nvidia docker container to get cuda on the github runner. The runners don't have gpus, so you can't test with them(at least without custom runners), but you can at least build them

[purepani]

Ok, I believe I have finally finished force pushing everything to get this to work fully.
Assuming your cuda stuff is correctly set up, you should be able to drop into the project folder and run uv run on anything to get the package using the previous PR; additionally any time you cut a release(with the version on the tag and the version in pyproject matching), it will upload to pypi once you set up the connection on pypi(this is pretty easy to do).

One thing I haven't done is test in a cuda environment, since I don't have an easily accessible one set up. If you can do that, that would be great(just installing the project into your environment with pip install . and testing it should work).

[purepani]

I just realized I completely forgot about windows support on the auto release
I'll work on that later

[AnderBiguri]

Thanks a lot @purepani , I'll try to test both things ASAP and then I'll merge. This is super appreciated, thanks!

[purepani]

Hmm I'm not sure if I have time in the near future to figure this out; will probably be at least early next week. Maybe it's better to just upload linux by itself(or upload windows manually with linux).
The issue with windows is I don't have a windows computer to test the build on, and also that I can't just run a docker image with cuda since docker is linux only, so I have to figure out how to install into the CI manually.

[purepani]

In researching on my phone while waiting for stuff, I found that there aren't super great solutions for windows cuda CI without custom runners, since pretty much all of them seem to be a bit manual, but they should work.
Mostly writing here as a note to myself for later.

[AnderBiguri]

thank @purepani . I am suddenly swamped with some stuff, so I will need to leave testing and mergin this for end of April, but I really want to do it, just letting you know you have time to faff around with this until then if you want :)

[purepani]

I'll definitely do that :)

[AnderBiguri]

Right! Finally tested and working on a linux machine with CUDA!

Anything else I should test before a merge?

Once merged, I'll try to figure out how to make a pip package then for tigre

[purepani]

I don't think you need to test anything else. Of course, a future step should be to get this working for windows though. I may or may not be able to get to it, but I do remember it was not tricky getting cuda toolkit installed on the GitHub runner. Maybe easiest to just use a conda environment for that. I may or may not be able to get to this in the future, but I've been procrastinating on my master's degree and need to get to finishing that for now. (Though feel free to ping me for questions!)

There is https://test.pypi.org to test the workflow before going on actual pypi. You'll need to create a trusted publisher, and link it to the GitHub actions, and the GitHub environment. The documentation for that is here: https://docs.pypi.org/trusted-publishers/
(The site in general is pretty comprehensive). Once you get it working on the test pypi, you can do the exact same thing on the actual pypi to do it!

The way the release workflow is triggered is creating a release, and have a branch named vX.Y.Z where X.Y.Z is the version.

You can also upload manually to pypi on your computer. Just build a wheel and upload it(I don't remember the pip commands off the top of my head, but the uv commands are uv build and uv publish).

[purepani]

Feel free to ping me with any questions!