As someone who has been far down this road and built libraries in other languages, I would espouse that this is mostly not worth time. The community for ML is

As someone who has been far down this road and built libraries in other languages, I would espouse that this is mostly not worth time. The community for ML is built around Python and C++, and you will be fighting upstream the whole way instead of just producing things that are useful to people. Cuda is written in C++ so most of everything builds up from that. Python has great C++ interop, not many languages do. The scale of tooling that has been built in Python is not easily replicated. I've watched libraries come and go over the years as they finally realize they can't keep up and it takes a big community to tackle these problems. You can learn a lot by writing ML in other languages, and I get it, Python leaves a lot to be desired, but if you want to build things that are truly useful, I would stick with the community.