Well-Read Students Learn Better: On the Importance of Pre-training Compact Models

TL;DR

This paper demonstrates that pre-training small models is crucial and competitive, and combining it with knowledge distillation from large models further enhances performance, providing a simple yet effective approach for compact NLP models.

Contribution

It highlights the importance of pre-training for small models and introduces Pre-trained Distillation, a straightforward method combining pre-training and knowledge distillation for improved compact models.

Findings

Pre-training small models remains important for performance.

Knowledge distillation from large models improves compact model accuracy.

Model size and data properties have a combined effect on results.

Abstract

Recent developments in natural language representations have been accompanied by large and expensive models that leverage vast amounts of general-domain text through self-supervised pre-training. Due to the cost of applying such models to down-stream tasks, several model compression techniques on pre-trained language representations have been proposed (Sun et al., 2019; Sanh, 2019). However, surprisingly, the simple baseline of just pre-training and fine-tuning compact models has been overlooked. In this paper, we first show that pre-training remains important in the context of smaller architectures, and fine-tuning pre-trained compact models can be competitive to more elaborate methods proposed in concurrent work. Starting with pre-trained compact models, we then explore transferring task knowledge from large fine-tuned models through standard knowledge distillation. The resulting simple, yet effective and general algorithm, Pre-trained Distillation, brings further improvements. Through extensive experiments, we more generally explore the interaction between pre-training and distillation under two variables that have been under-studied: model size and properties of unlabeled task data. One surprising observation is that they have a compound effect even when sequentially applied on the same data. To accelerate future research, we will make our 24 pre-trained miniature BERT models publicly available.