Great news but I'd like to know how does it compare with just using torchlib. If this is faster than torchlib this optimizations should flow to torchlib as wel

Great news but I'd like to know how does it compare with just using torchlib. If this is faster than torchlib this optimizations should flow to torchlib as well Love the idea of not having to deal with python, though; dependency management is just horrible, I'd much rather have ML projects written in cpp.