Conda packages for cuequivariance-ops-*

I was wondering if you would consider creating a conda package for the cuequivariance-ops-(torch|jax)-cu(12|13) libraries. Optimally, for me would be if they could also be published on conda-forge, but that might be difficult given the license. Maybe there is another public channel that could be used.

I am currently part of a team that is trying to create a conda packaged version of the OpenFold-3 model. Having all the cuequivariance dependencies in conda(-forge) would help us immensely in this endeavour. The base cuequivariance libraries are already available on conda-forge. It is only the -ops libraries that are missing.

I was also trying to and would be happy to create the packages myself, but at least with grayskull this turned out to be difficult as there is (intentionally) no source distribution published on PyPI.

Of course, as I do not have access to the source, I do not know how difficult the build process for such a package would be. At the same time, given that the libraries have been packaged for PyPI, I would hope that the translation is rather straight forward (e.g., with the help of grayskull). Naturally, I completely understand if this is not a priority at this time.

Thank you for the library and your consideration 🙂

Start by inspecting the PyPI packages for cuequivariance-ops-(torch|jax)-cu(12|13), the existing conda-forge packages for the base libraries, and grayskull's output. Check whether the package artifacts and license permit a conda recipe or another public channel. Done means the requested Torch and JAX CUDA 12 and 13 variants are buildable and published through an agreed channel.