FastBERT: a Self-distilling BERT with Adaptive Inference Time
This addresses the problem of high computational costs for deploying BERT in resource-limited scenarios, offering a tunable solution for efficiency.
The paper tackles the computational inefficiency of BERT models by proposing FastBERT, a self-distilling model with adaptive inference time, which achieves speedups of 1 to 12 times over BERT on twelve datasets with minimal performance loss.
Pre-trained language models like BERT have proven to be highly performant. However, they are often computationally expensive in many practical scenarios, for such heavy models can hardly be readily implemented with limited resources. To improve their efficiency with an assured model performance, we propose a novel speed-tunable FastBERT with adaptive inference time. The speed at inference can be flexibly adjusted under varying demands, while redundant calculation of samples is avoided. Moreover, this model adopts a unique self-distillation mechanism at fine-tuning, further enabling a greater computational efficacy with minimal loss in performance. Our model achieves promising results in twelve English and Chinese datasets. It is able to speed up by a wide range from 1 to 12 times than BERT if given different speedup thresholds to make a speed-performance tradeoff.