PyTorch is one of the main tools used for machine learning research these days. It’s been developed in beta mode for over 2 years, but this October, a release candidate for 1.0 version has been finally released! In this talk, I’ll briefly introduce the library, and then move on to showcase the cutting edge features we introduced recently.
The talk will be divided into multiple sections. First, an extremely quick introduction to what PyTorch is, and what can it be used for (including use cases outside of machine learning!). Then, I will cover a number of topics that are interesting in the current context of the library, including: - Hybrid frontend (JIT compiler) - Path from research to production - C++ API and inference - Caffe2 merger - New distributed backend