Yeah, can anybody explain why the development is stalled?

It does not look like it.
https://medium.com/@mjhirn/tensorflow-wins-89b78b29aafb

Ok, it appears Leaf is still maintained under a different name: https://github.com/spearow/juice

is nobody using this? This is a bit weird, seems like more and more people would be interested in a Rust CUDA framework

What's the reason for the fork? I guess autumnai just stopped the development?

"Now that good-enough tools, to build maintainable machine-learning applications, like Tensorflow and Keras, exist, venture capital flows more and more into companies who try to create immense value in verticals with new AI-driven applications, instead of infrastructure providers."

I'm here looking into leaf because I don't feel tensorflow is good enough. I'm fed up with the nondeterminism, API churn caused by adopting keras (splitting the keras using community) and now the 2.0 migration, the exploding complexity of the codebase, the flood of warnings, the long startup time, JIT screwing up benchmarks, the pervasive use of hidden state in the API, the constant battle to keep it working with your GPU driver, tf reserving ALL of your gpu memory by default,...

If anyone out there is reading this, check out this new deep learning framework in Rust called Burn: https://github.com/burn-rs/burn