PyTorch New Quantization API

A brief summary why PyTorch has a new prototype API for quantization - looks like the previous API was too difficult? It wasn't really, but it required some fiddling and non-standard layers just did not work: