Video lesson 6

Building makemore Part 5: Building a WaveNet

Turn the Module 4 toolkit into a real architecture. Torch-ify the model into one Sequential object, stretch the context from 3 characters to 8, then rebuild the network as a WaveNet-style hierarchy that fuses characters two at a time - and hunt down a real bug that has been hiding in your own BatchNorm the whole time. The deepest network of the course trains to the best validation loss of the course.