GTC 2020: Named Tensors, Model Quantization, and the Latest PyTorch Features

GTC 2020 S22145
Presenters: Joseph Spisak,Facebook; James Reed,Facebook AI
Abstract
PyTorch, the popular open-source ML framework, has continued to evolve rapidly since the introduction of PyTorch 1.0, which brought an accelerated workflow from research to production. We’ll deep dive on some of the most important new advances, including the ability to name tensors, support for quantization-aware training and post-training quantization, improved distributed training on GPUs, and streamlined mobile deployment. We’ll also cover new developer tools and domain-specific frameworks including Captum for model interpretability, Detectron2 for computer vision, and speech extensions for Fairseq.

Watch this session
Join in the conversation below.