https://bugzilla.redhat.com/show_bug.cgi?id=2242971
--- Comment #34 from Tom Rix trix@redhat.com --- (In reply to Sandro from comment #33)
A minor thing that caught my eye: the title of the review request should be of the form 'Review Request: <main package name here> - <short summary here>'. The '<short summary here>' is usually the %{summary} from the spec file. Looking at the spec file, that would be "An AI/ML python package". However, on PyPI I see "Tensors and Dynamic neural networks in Python with strong GPU acceleration". That's probably too long for the spec file summary, yet more to the point, I guess (except for the GPU acceleration part).
Just a shout from the side line. Feel free to ignore.
PyTorch is such a big and well know project, I think of more as a brand name like 'Coke' , that's why i added the '- PyTorch' to this review instead of the short summary in the spec, which as you say isn't the long thing in the git project because it is long and at this point misleading wrt GPU acceleration.
CUDA is impossible. ROCm is planned. Sycl/OneAPI would be nice, but at this time unplanned. Vulkan and OpenCL I do not think have full support in PyTorch, missing some/most of the necessary math libs.