BiT: Robustly Binarized Multi-distilled Transformer

TL;DR

This paper introduces BiT, a set of techniques enabling fully binarized transformer models that maintain high accuracy, making them practical for resource-limited environments, by combining innovative binarization, activation, and distillation methods.

Contribution

The paper presents a novel combination of binarization schemes, elastic binary activation functions, and progressive distillation to create highly accurate fully binarized transformers.

Findings

Achieves near full-precision BERT accuracy on GLUE benchmark

Introduces a two-set binarization scheme and elastic binary activation

Demonstrates effective model distillation to limit precision without large accuracy loss

Abstract

Modern pre-trained transformers have rapidly advanced the state-of-the-art in machine learning, but have also grown in parameters and computational complexity, making them increasingly difficult to deploy in resource-constrained environments. Binarization of the weights and activations of the network can significantly alleviate these issues, however, is technically challenging from an optimization perspective. In this work, we identify a series of improvements that enables binary transformers at a much higher accuracy than what was possible previously. These include a two-set binarization scheme, a novel elastic binary activation function with learned parameters, and a method to quantize a network to its limit by successively distilling higher precision models into lower precision students. These approaches allow for the first time, fully binarized transformer models that are at a practical level of accuracy, approaching a full-precision BERT baseline on the GLUE language understanding benchmark within as little as 5.9%. Code and models are available at: https://github.com/facebookresearch/bit.