Abstract
Knowledge distillation (KD) involves training a small “student” model to replicate the strong performance of a high-capacity “teacher” model, enabling efficient deployment in resource-constrained settings. Top-performing methods tend to be task- or architecture-specific and lack generalizability. Several existing approaches require pretraining of the teacher on task-specific datasets, which can be costly for large and unstable for small datasets. Here we propose an approach for improving KD through a novel distillation loss agnostic to the task and model architecture. We successfully apply our method to the distillation of the BERT-base and achieve highly competitive results from the distilled student across a range of GLUE tasks, especially for tasks with smaller datasets.- Anthology ID:
- 2023.findings-acl.463
- Volume:
- Findings of the Association for Computational Linguistics: ACL 2023
- Month:
- July
- Year:
- 2023
- Address:
- Toronto, Canada
- Editors:
- Anna Rogers, Jordan Boyd-Graber, Naoaki Okazaki
- Venue:
- Findings
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 7346–7354
- Language:
- URL:
- https://aclanthology.org/2023.findings-acl.463
- DOI:
- 10.18653/v1/2023.findings-acl.463
- Cite (ACL):
- Sayantan Dasgupta, Trevor Cohn, and Timothy Baldwin. 2023. Cost-effective Distillation of Large Language Models. In Findings of the Association for Computational Linguistics: ACL 2023, pages 7346–7354, Toronto, Canada. Association for Computational Linguistics.
- Cite (Informal):
- Cost-effective Distillation of Large Language Models (Dasgupta et al., Findings 2023)
- PDF:
- https://preview.aclanthology.org/nschneid-patch-2/2023.findings-acl.463.pdf