Distilling Linguistic Context for Language Model Compression

TL;DR

This paper introduces a novel knowledge distillation method that transfers contextual relationships in language representations, improving language model compression without architectural restrictions, validated across diverse benchmarks and models.

Contribution

It proposes a new distillation objective focusing on relational knowledge transfer, enabling flexible architecture adaptation and enhanced language understanding performance.

Findings

Effective across various model sizes and architectures

Improves performance on language understanding benchmarks

Compatible with adaptive size pruning methods like DynaBERT

Abstract

A computationally expensive and memory intensive neural network lies behind the recent success of language representation learning. Knowledge distillation, a major technique for deploying such a vast language model in resource-scarce environments, transfers the knowledge on individual word representations learned without restrictions. In this paper, inspired by the recent observations that language representations are relatively positioned and have more semantic knowledge as a whole, we present a new knowledge distillation objective for language representation learning that transfers the contextual knowledge via two types of relationships across representations: Word Relation and Layer Transforming Relation. Unlike other recent distillation techniques for the language models, our contextual distillation does not have any restrictions on architectural changes between teacher and student. We validate the effectiveness of our method on challenging benchmarks of language understanding tasks, not only in architectures of various sizes, but also in combination with DynaBERT, the recently proposed adaptive size pruning method.