Learning Pytorch while implementing Wavenet worked quite well!!
Work on recreating Wavenet has now moved to a private reposetory for work reasons. The private version has CUDA support, parallelisation over multiple GPUs, its own batch generator, and separated files for the Utils, model, and training, plus other optimisations.
Work will continue to go ahead on the private version of global conditioning, caching for faster prediction, and further optimisation.
Contact: Riccardo Grinover - grinovericcardo@gmail.com