NarrowBERT: Accelerating Masked Language Model Pretraining and Inference

TL;DR

NarrowBERT is a sparsified transformer model that significantly accelerates masked language model pretraining and inference by focusing computation on masked tokens, with minimal impact on downstream task performance.

Contribution

It introduces a novel sparsification technique for transformers that boosts pretraining and inference throughput without degrading accuracy.

Findings

Over 2x faster pretraining throughput

Up to 3.5x faster inference speed

Comparable performance on downstream NLP tasks

Abstract

Large-scale language model pretraining is a very successful form of self-supervised learning in natural language processing, but it is increasingly expensive to perform as the models and pretraining corpora have become larger over time. We propose NarrowBERT, a modified transformer encoder that increases the throughput for masked language model pretraining by more than $2\times$. NarrowBERT sparsifies the transformer model such that the self-attention queries and feedforward layers only operate on the masked tokens of each sentence during pretraining, rather than all of the tokens as with the usual transformer encoder. We also show that NarrowBERT increases the throughput at inference time by as much as $3.5\times$ with minimal (or no) performance degradation on sentence encoding tasks like MNLI. Finally, we examine the performance of NarrowBERT on the IMDB and Amazon reviews classification and CoNLL NER tasks and show that it is also comparable to standard BERT performance.