(Update Aug. 3 2025)

Another fallen soldier. Even more noble than its SciPy/NumPy predecessor, this PyTorch implementation correctly computes JVPs and VJPs. Alas, it is too slow. Still, many lessons were gleaned from this effort and its carcass can still be scavenged for some useful code (albeit one has to battle through poor abstractions as the abstract/final pattern was not used for this implementation).