BERT-of-Theseus: Compressing BERT by Progressive Module Replacing
This work addresses the problem of reducing BERT's computational cost for NLP practitioners, offering a novel compression technique that is incremental in improving efficiency.
The paper tackles BERT model compression by introducing a progressive module replacing method that trains compact substitutes to mimic original modules without extra loss functions, achieving superior performance on the GLUE benchmark compared to prior knowledge distillation approaches.
In this paper, we propose a novel model compression approach to effectively compress BERT by progressive module replacing. Our approach first divides the original BERT into several modules and builds their compact substitutes. Then, we randomly replace the original modules with their substitutes to train the compact modules to mimic the behavior of the original modules. We progressively increase the probability of replacement through the training. In this way, our approach brings a deeper level of interaction between the original and compact models. Compared to the previous knowledge distillation approaches for BERT compression, our approach does not introduce any additional loss function. Our approach outperforms existing knowledge distillation approaches on GLUE benchmark, showing a new perspective of model compression.