This is a remarkable, very unusual project. It looks well thought out and is surprisingly complete -- it even includes an end-to-end ML training and sampling program as an example!
Is there a writeup somewhere that explains more about the motivation and history for this project?
