Abstract
In this work, we empirically show that updating pretrained LMs (350M, 1.3B, 2.7B) with just a few steps of Gradient Ascent Post-training (GAP) on random, unlabeled text corpora enhances its zero-shot generalization capabilities across diverse NLP tasks. Specifically, we show that GAP can allow LMs to become comparable to 2-3x times larger LMs across 12 different NLP tasks. We also show that applying GAP on out-of-distribution corpora leads to the most reliable performance improvements. Our findings indicate that GAP can be a promising method for improving the generalization capability of LMs without any task-specific fine-tuning.- Anthology ID:
- 2023.acl-short.74
- Volume:
- Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)
- Month:
- July
- Year:
- 2023
- Address:
- Toronto, Canada
- Editors:
- Anna Rogers, Jordan Boyd-Graber, Naoaki Okazaki
- Venue:
- ACL
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 851–864
- Language:
- URL:
- https://aclanthology.org/2023.acl-short.74
- DOI:
- 10.18653/v1/2023.acl-short.74
- Cite (ACL):
- Dongkeun Yoon, Joel Jang, Sungdong Kim, and Minjoon Seo. 2023. Gradient Ascent Post-training Enhances Language Model Generalization. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 851–864, Toronto, Canada. Association for Computational Linguistics.
- Cite (Informal):
- Gradient Ascent Post-training Enhances Language Model Generalization (Yoon et al., ACL 2023)
- PDF:
- https://preview.aclanthology.org/ingest-acl-2023-videos/2023.acl-short.74.pdf