Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities

Baolong Bi, Shenghua Liu, Yiwei Wang, Lingrui Mei, Hongcheng Gao, Yilong Xu, Xueqi Cheng


Abstract
The parametric knowledge memorized by large language models (LLMs) becomes outdated quickly. In-context editing (ICE) is currently the most effective method for updating the knowledge of LLMs. Recent advancements involve enhancing ICE by modifying the decoding strategy, obviating the need for altering internal model structures or adjusting external prompts.However, this enhancement operates across the entire sequence generation, encompassing a plethora of non-critical tokens.In this work, we introduce **A**daptive **T**oken **Bias**er (ATBias), a new decoding technique designed to enhance ICE.It focuses on the tokens that are mostly related to knowledge during decoding, biasing their logits by matching key entities related to new and parametric knowledge.Experimental results show that ATBias significantly enhances ICE performance, achieving up to a 32.3% improvement over state-of-the-art ICE methods while incurring only half the latency.ATBias not only improves the knowledge editing capabilities of ICE but can also be widely applied to LLMs with negligible cost.
Anthology ID:
2024.findings-emnlp.647
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2024
Month:
November
Year:
2024
Address:
Miami, Florida, USA
Editors:
Yaser Al-Onaizan, Mohit Bansal, Yun-Nung Chen
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
11071–11083
Language:
URL:
https://preview.aclanthology.org/build-pipeline-with-new-library/2024.findings-emnlp.647/
DOI:
10.18653/v1/2024.findings-emnlp.647
Bibkey:
Cite (ACL):
Baolong Bi, Shenghua Liu, Yiwei Wang, Lingrui Mei, Hongcheng Gao, Yilong Xu, and Xueqi Cheng. 2024. Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 11071–11083, Miami, Florida, USA. Association for Computational Linguistics.
Cite (Informal):
Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities (Bi et al., Findings 2024)
Copy Citation:
PDF:
https://preview.aclanthology.org/build-pipeline-with-new-library/2024.findings-emnlp.647.pdf