Neural Network Distiller: A PyTorch environment for neural network compression

Description

Deep learning (DL) and artificial intelligence (AI) are quickly becoming ubiquitous. DL applications employ deep neural networks (DNNs), which are notoriously time, compute, energy, and memory intensive.

Intel’s AI Lab has recently open-sourced Neural Network Distiller, a Python package for neural network compression research. Distiller provides a PyTorch environment for prototyping and analyzing compression algorithms, such as sparsity-inducing methods and low-precision arithmetic. Intel AI thinks that DNN compression can be another catalyst that will help bring deep learning innovation to more industries and application domains, making our lives easier, healthier, and more productive.

Distiller is built with the following features and tools, keeping both DL researchers and engineers in mind:

• A framework for integrating pruning, regularization, and quantization algorithms
• A set of tools for analyzing and evaluating compression performance
• Example implementations of state-of-the-art compression algorithms

Neta Zmora discusses the motivation for compressing DNNs, outlines compression approaches, and explores Distiller’s design and tools, supported algorithms, and code and documentation. Neta concludes with an example implementation of a compression research paper.

For more information on Distiller, check out Intel AI’s introductory blog post.