Accelerating Training of Transformer-Based Language Models with Progressive Layer Dropping

  • Minjia Zhang ,
  • Yuxiong He

34th Conference on Neural Information Processing Systems (NeurIPS 2020) |

Recently, Transformer-based language models have demonstrated remarkable performance across many NLP domains. However, the unsupervised pre-training step of these models suffers from unbearable overall computational expenses. Current methods for accelerating the pre-training either rely on massive parallelism with advanced hardware or are not applicable to language modeling. In this work, we propose a method based on progressive layer dropping that speeds the training of Transformer-based language models, not at the cost of excessive hardware resources but from model architecture change and training technique boosted efficiency. Extensive experiments on BERT show that the proposed method achieves  a 24% time reduction on average per sample and allows the pre-training to be 2.5X faster than the baseline to get a similar accuracy on downstream tasks. While being faster, our pre-trained models are equipped with strong knowledge transferability, achieving a 1.1 point higher GLUE score than the baseline when pre-trained with the same number of samples.