DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

• DOI: 10.48550/arxiv.1910.01108
• Publication Date: 2019-01-01
• Authors: Sanh, Victor; Debut, Lysandre; Chaumond, Julien; Wolf, Thomas
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Computation and Language (cs.CL)

February 2020 - Revision: fix bug in evaluation metrics, updated metrics, argumentation unchanged. 5 pages, 1 figure, 4 tables. Accepted at the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing - NeurIPS 2019<br/>As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP), operating these large models in on-the-edge and/or under constrained computational training or inference budgets remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage knowledge distillation during the pre-training phase and show that it is possible to reduce the size of a BERT model by 40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive biases learned by larger models during pre-training, we introduce a triple loss combining language modeling, distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device study.<br/>