The Sockeye 2 Neural Machine Translation Toolkit at AMTA 2020

TL;DR

Sockeye 2 is an improved neural machine translation toolkit that offers faster training and inference, higher accuracy, and easier deployment through modern features like MXNet Gluon API and quantization.

Contribution

The paper introduces Sockeye 2, a streamlined NMT toolkit with modern architecture, mixed precision training, and efficient decoding, enhancing performance and usability.

Findings

Faster training and inference times.

Higher automatic metric scores.

Efficient CPU decoding with 8-bit quantization.

Abstract

We present Sockeye 2, a modernized and streamlined version of the Sockeye neural machine translation (NMT) toolkit. New features include a simplified code base through the use of MXNet's Gluon API, a focus on state of the art model architectures, distributed mixed precision training, and efficient CPU decoding with 8-bit quantization. These improvements result in faster training and inference, higher automatic metric scores, and a shorter path from research to production.