Conventional wisdom in pruning Transformer-based language models is that pruning reduces the model expressiveness and thus is more likely to underfit rather than overfit. However, under the trending pretrain-and-finetune paradigm, we postulate a counter-traditional hypothesis, that is: pruning increases the risk of overfitting when performed at the fine-tuning phase. In this paper, we aim to address the overfitting problem and improve pruning performance via progressive knowledge distillation with error-bound properties. We show for the first time that reducing the risk of overfitting can help the effectiveness of pruning under the pretrain-and-finetune paradigm. Ablation studies and experiments on the GLUE benchmark show that our method outperforms the leading competitors across different tasks.
@article{arxiv.2110.08190,
title = {Sparse Progressive Distillation: Resolving Overfitting under Pretrain-and-Finetune Paradigm},
author = {Shaoyi Huang and Dongkuan Xu and Ian E. H. Yen and Yijue Wang and Sung-en Chang and Bingbing Li and Shiyang Chen and Mimi Xie and Sanguthevar Rajasekaran and Hang Liu and Caiwen Ding},
journal= {arXiv preprint arXiv:2110.08190},
year = {2023}
}
Comments
11 pages; 16 figures; Published in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing