BinaryBERT: Pushing the Limit of BERT Quantization
This work addresses the problem of deploying large language models on resource-constrained devices for practitioners by drastically reducing model size.
This paper introduces BinaryBERT, a method that binarizes BERT's weights to achieve significant model compression. By using ternary weight splitting for initialization, BinaryBERT achieves a 24x smaller model size with only a slight performance drop compared to full-precision models, setting new state-of-the-art compression results on GLUE and SQuAD benchmarks.
The rapid development of large pre-trained language models has greatly increased the demand for model compression techniques, among which quantization is a popular solution. In this paper, we propose BinaryBERT, which pushes BERT quantization to the limit by weight binarization. We find that a binary BERT is hard to be trained directly than a ternary counterpart due to its complex and irregular loss landscape. Therefore, we propose ternary weight splitting, which initializes BinaryBERT by equivalently splitting from a half-sized ternary network. The binary model thus inherits the good performance of the ternary one, and can be further enhanced by fine-tuning the new architecture after splitting. Empirical results show that our BinaryBERT has only a slight performance drop compared with the full-precision model while being 24x smaller, achieving the state-of-the-art compression results on the GLUE and SQuAD benchmarks.