On the Transformer Growth for Progressive BERT Training

Xiaotao Gu, Liyuan Liu, Hongkun Yu, Jing Li, Chen Chen, Jiawei Han


Abstract
As the excessive pre-training cost arouses the need to improve efficiency, considerable efforts have been made to train BERT progressively–start from an inferior but low-cost model and gradually increase the computational complexity. Our objective is to help advance the understanding of such Transformer growth and discover principles that guide progressive training. First, we find that similar to network architecture selection, Transformer growth also favors compound scaling. Specifically, while existing methods only conduct network growth in a single dimension, we observe that it is beneficial to use compound growth operators and balance multiple dimensions (e.g., depth, width, and input length of the model). Moreover, we explore alternative growth operators in each dimension via controlled comparison to give practical guidance for operator selection. In light of our analyses, the proposed method CompoundGrow speeds up BERT pre-training by 73.6% and 82.2% for the base and large models respectively while achieving comparable performances.
Anthology ID:
2021.naacl-main.406
Volume:
Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies
Month:
June
Year:
2021
Address:
Online
Editors:
Kristina Toutanova, Anna Rumshisky, Luke Zettlemoyer, Dilek Hakkani-Tur, Iz Beltagy, Steven Bethard, Ryan Cotterell, Tanmoy Chakraborty, Yichao Zhou
Venue:
NAACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
5174–5180
Language:
URL:
https://aclanthology.org/2021.naacl-main.406
DOI:
10.18653/v1/2021.naacl-main.406
Bibkey:
Cite (ACL):
Xiaotao Gu, Liyuan Liu, Hongkun Yu, Jing Li, Chen Chen, and Jiawei Han. 2021. On the Transformer Growth for Progressive BERT Training. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5174–5180, Online. Association for Computational Linguistics.
Cite (Informal):
On the Transformer Growth for Progressive BERT Training (Gu et al., NAACL 2021)
Copy Citation:
PDF:
https://preview.aclanthology.org/nschneid-patch-4/2021.naacl-main.406.pdf
Video:
 https://preview.aclanthology.org/nschneid-patch-4/2021.naacl-main.406.mp4
Data
GLUE