TY - GEN
T1 - Length-adaptive transformer
T2 - Joint Conference of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, ACL-IJCNLP 2021
AU - Kim, Gyuwan
AU - Cho, Kyunghyun
N1 - Funding Information:
The authors appreciate Clova AI members and the anonymous reviewers for their constructive feedback. Specifically, Dongyoon Han and Byeongho Heo introduced relevant works and gave insights from the view of the computer vision community. We use Naver Smart Machine Learning (Sung et al., 2017; Kim et al., 2018) platform for the experiments.
Publisher Copyright:
© 2021 Association for Computational Linguistics
PY - 2021
Y1 - 2021
N2 - Despite transformers' impressive accuracy, their computational cost is often prohibitive to use with limited computational resources. Most previous approaches to improve inference efficiency require a separate model for each possible computational budget. In this paper, we extend PoWER-BERT (Goyal et al., 2020) and propose Length-Adaptive Transformer that can be used for various inference scenarios after one-shot training. We train a transformer with LengthDrop, a structural variant of dropout, which stochastically determines a sequence length at each layer. We then conduct a multi-objective evolutionary search to find a length configuration that maximizes the accuracy and minimizes the efficiency metric under any given computational budget. Additionally, we significantly extend the applicability of PoWER-BERT beyond sequence-level classification into token-level classification with Drop-and-Restore process that drops word-vectors temporarily in intermediate layers and restores at the last layer if necessary. We empirically verify the utility of the proposed approach by demonstrating the superior accuracy-efficiency trade-off under various setups, including span-based question answering and text classification. Code is available at https://github.com/clovaai/length-adaptive-transformer.
AB - Despite transformers' impressive accuracy, their computational cost is often prohibitive to use with limited computational resources. Most previous approaches to improve inference efficiency require a separate model for each possible computational budget. In this paper, we extend PoWER-BERT (Goyal et al., 2020) and propose Length-Adaptive Transformer that can be used for various inference scenarios after one-shot training. We train a transformer with LengthDrop, a structural variant of dropout, which stochastically determines a sequence length at each layer. We then conduct a multi-objective evolutionary search to find a length configuration that maximizes the accuracy and minimizes the efficiency metric under any given computational budget. Additionally, we significantly extend the applicability of PoWER-BERT beyond sequence-level classification into token-level classification with Drop-and-Restore process that drops word-vectors temporarily in intermediate layers and restores at the last layer if necessary. We empirically verify the utility of the proposed approach by demonstrating the superior accuracy-efficiency trade-off under various setups, including span-based question answering and text classification. Code is available at https://github.com/clovaai/length-adaptive-transformer.
UR - http://www.scopus.com/inward/record.url?scp=85117464537&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85117464537&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85117464537
T3 - ACL-IJCNLP 2021 - 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, Proceedings of the Conference
SP - 6501
EP - 6511
BT - ACL-IJCNLP 2021 - 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, Proceedings of the Conference
PB - Association for Computational Linguistics (ACL)
Y2 - 1 August 2021 through 6 August 2021
ER -