PyTorch* 2.0 Overview & Technical Deep Dive into Its Deep Learning Compiler

Initially released by Facebook* (now Meta*) in the fall of 2016, PyTorch* has become one of the most popular deep learning frameworks for compute-intensive training and inference.

This session focuses on PyTorch 2.0. Released on March 15, 2023, it offers the same eager-mode development and user experience while fundamentally changing and supercharging how PyTorch operates at the compiler level.

Hear engineers from Meta and Intel discuss:

• What’s new in PyTorch 2.0.
• Its deep learning compiler stack.
• TorchInductor, which is the deep learning compiler back end that supports training and multiple back-end targets.
• Features Intel contributed to the new release, such as technologies from Intel® Extension for PyTorch* and int8 inference optimizations provided by Intel® oneAPI Deep Neural Network Library (oneDNN).

They also provide a demo.

Skill level: Intermediate

• Get the Intel Extension for PyTorch from GitHub*.
• Get the stand-alone version of oneDNN or as part of the Intel® oneAPI Base Toolkit.