Revisiting Intermediate Layer Distillation for Compressing Language Models: An Overfitting Perspective

TL;DR

This paper identifies overfitting issues in intermediate layer distillation for language models and proposes a regularized method that improves performance on benchmarks by focusing on the last layer and auxiliary tasks.

Contribution

It introduces a simple consistency-regularized ILD method that mitigates overfitting and enhances knowledge distillation effectiveness for language models.

Findings

CR-ILD outperforms existing KD methods on GLUE benchmark

Distilling only the last Transformer layer improves generalization

Using supplementary tasks reduces overfitting in ILD

Abstract

Knowledge distillation (KD) is a highly promising method for mitigating the computational problems of pre-trained language models (PLMs). Among various KD approaches, Intermediate Layer Distillation (ILD) has been a de facto standard KD method with its performance efficacy in the NLP field. In this paper, we find that existing ILD methods are prone to overfitting to training datasets, although these methods transfer more information than the original KD. Next, we present the simple observations to mitigate the overfitting of ILD: distilling only the last Transformer layer and conducting ILD on supplementary tasks. Based on our two findings, we propose a simple yet effective consistency-regularized ILD (CR-ILD), which prevents the student model from overfitting the training dataset. Substantial experiments on distilling BERT on the GLUE benchmark and several synthetic datasets demonstrate that our proposed ILD method outperforms other KD techniques. Our code is available at https://github.com/jongwooko/CR-ILD.