Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wu, Taiqiang, Hou, Cheng, Lao, Shanshan, Li, Jiayi, Wong, Ngai, Zhao, Zhe, Yang, Yujiu
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2305.09098
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910375364001792
author Wu, Taiqiang
Hou, Cheng
Lao, Shanshan
Li, Jiayi
Wong, Ngai
Zhao, Zhe
Yang, Yujiu
author_facet Wu, Taiqiang
Hou, Cheng
Lao, Shanshan
Li, Jiayi
Wong, Ngai
Zhao, Zhe
Yang, Yujiu
contents Knowledge Distillation (KD) is a predominant approach for BERT compression. Previous KD-based methods focus on designing extra alignment losses for the student model to mimic the behavior of the teacher model. These methods transfer the knowledge in an indirect way. In this paper, we propose a novel Weight-Inherited Distillation (WID), which directly transfers knowledge from the teacher. WID does not require any additional alignment loss and trains a compact student by inheriting the weights, showing a new perspective of knowledge distillation. Specifically, we design the row compactors and column compactors as mappings and then compress the weights via structural re-parameterization. Experimental results on the GLUE and SQuAD benchmarks show that WID outperforms previous state-of-the-art KD-based baselines. Further analysis indicates that WID can also learn the attention patterns from the teacher model without any alignment loss on attention distributions. The code is available at https://github.com/wutaiqiang/WID-NAACL2024.
format Preprint
id arxiv_https___arxiv_org_abs_2305_09098
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Weight-Inherited Distillation for Task-Agnostic BERT Compression
Wu, Taiqiang
Hou, Cheng
Lao, Shanshan
Li, Jiayi
Wong, Ngai
Zhao, Zhe
Yang, Yujiu
Computation and Language
Machine Learning
Knowledge Distillation (KD) is a predominant approach for BERT compression. Previous KD-based methods focus on designing extra alignment losses for the student model to mimic the behavior of the teacher model. These methods transfer the knowledge in an indirect way. In this paper, we propose a novel Weight-Inherited Distillation (WID), which directly transfers knowledge from the teacher. WID does not require any additional alignment loss and trains a compact student by inheriting the weights, showing a new perspective of knowledge distillation. Specifically, we design the row compactors and column compactors as mappings and then compress the weights via structural re-parameterization. Experimental results on the GLUE and SQuAD benchmarks show that WID outperforms previous state-of-the-art KD-based baselines. Further analysis indicates that WID can also learn the attention patterns from the teacher model without any alignment loss on attention distributions. The code is available at https://github.com/wutaiqiang/WID-NAACL2024.
title Weight-Inherited Distillation for Task-Agnostic BERT Compression
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2305.09098