A Study on Knowledge Distillation from Weak Teacher for Scaling Up Pre-trained Language Models
Hayeon Lee, Rui Hou, Jongpil Kim, Davis Liang, Sung Ju Hwang, Alexander Min
Abstract
Distillation from Weak Teacher (DWT) is a method of transferring knowledge from a smaller, weaker teacher model to a larger student model to improve its performance. Previous studies have shown that DWT can be effective in the vision domain and natural language processing (NLP) pre-training stage. Specifically, DWT shows promise in practical scenarios, such as enhancing new generation or larger models using pre-trained yet older or smaller models and lacking a resource budget. However, the optimal conditions for using DWT have yet to be fully investigated in NLP pre-training. Therefore, this study examines three key factors to optimize DWT, distinct from those used in the vision domain or traditional knowledge distillation. These factors are:(i) the impact of teacher model quality on DWT effectiveness, (ii) guidelines for adjusting the weighting value for DWT loss, and (iii) the impact of parameter remapping as a student model initialization technique for DWT.
BibTeX
@inproceedings{lee-etal-2023-study,
title = "A Study on Knowledge Distillation from Weak Teacher for Scaling Up Pre-trained Language Models",
author = "Lee, Hayeon and
Hou, Rui and
Kim, Jongpil and
Liang, Davis and
Hwang, Sung Ju and
Min, Alexander",
editor = "Rogers, Anna and
Boyd-Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.findings-acl.714/",
doi = "10.18653/v1/2023.findings-acl.714",
pages = "11239--11246"
}