Non-Autoregressive Machine Translation: It's Not as Fast as it Seems

TL;DR

This paper critically evaluates non-autoregressive machine translation models, revealing that they are not consistently faster than autoregressive models in realistic scenarios and emphasizing the need for fair, comprehensive evaluation methods.

Contribution

It highlights flaws in current evaluation practices for NAR models and provides a fair comparison with autoregressive models under realistic conditions.

Findings

NAR models are faster on GPUs with small batch sizes.

Under realistic conditions, NAR models are often slower than AR models.

Calls for standardized and extensive evaluation of NAR models.

Abstract

Efficient machine translation models are commercially important as they can increase inference speeds, and reduce costs and carbon emissions. Recently, there has been much interest in non-autoregressive (NAR) models, which promise faster translation. In parallel to the research on NAR models, there have been successful attempts to create optimized autoregressive models as part of the WMT shared task on efficient translation. In this paper, we point out flaws in the evaluation methodology present in the literature on NAR models and we provide a fair comparison between a state-of-the-art NAR model and the autoregressive submissions to the shared task. We make the case for consistent evaluation of NAR models, and also for the importance of comparing NAR models with other widely used methods for improving efficiency. We run experiments with a connectionist-temporal-classification-based (CTC) NAR model implemented in C++ and compare it with AR models using wall clock times. Our results show that, although NAR models are faster on GPUs, with small batch sizes, they are almost always slower under more realistic usage conditions. We call for more realistic and extensive evaluation of NAR models in future work.