Revisiting Intermediate Layer Distillation for Compressing Language Models: An Overfitting Perspective
This work tackles computational inefficiency in pre-trained language models for NLP applications, presenting an incremental improvement to existing distillation methods.
The paper addresses overfitting in Intermediate Layer Distillation (ILD) for compressing language models, proposing a consistency-regularized method that outperforms other techniques on BERT distillation across GLUE and synthetic datasets.
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.