InstructEd: Soft-Instruction Tuning for Model Editing with Hops

Page view(s)
9
Checked on Nov 19, 2024
InstructEd: Soft-Instruction Tuning for Model Editing with Hops
Title:
InstructEd: Soft-Instruction Tuning for Model Editing with Hops
Journal Title:
Findings of the Association for Computational Linguistics ACL 2024
DOI:
Keywords:
Publication Date:
16 August 2024
Citation:
XiaoQi Han, Ru Li, Xiaoli Li, Jiye Liang, Zifang Zhang, and Jeff Pan. 2024. InstructEd: Soft-Instruction Tuning for Model Editing with Hops. In Findings of the Association for Computational Linguistics ACL 2024, pages 14953–14968, Bangkok, Thailand and virtual meeting. Association for Computational Linguistics.
Abstract:
The task of model editing becomes popular for correcting inaccurate or outdated parametric knowledge in Large Language Models (LLMs). However, there are major limitations of state of the art (SOTA) model editing methods, including the excessive memorization issue caused by the direct editing methods, as well as the error propagation and knowledge conflict issues from the memory enhancement methods, resulting in hindering models’ portability, e.g., the ability to transfer the new knowledge to related one-hop or multi-hop content. To address these issues, we propose the InstructEd method, the idea of which is to insert soft instructions into the attention module so as to facilitate interactions between instructions and questions and to understand and utilize new facts. Our main findings are: (i) InstructEd has achieved SOTA performance on three datasets for onehop/ multi-hop evaluation with LLaMAs and GPT2, achieving 10% (5%) improvement in one-hop (multi-hop) model editing. (ii) Different from earlier methods on editing parameters in FFN, we show that editing attention can also help. (iii) Model editing is highly related to retrieval augmented methods, which can help improve the locality of model editing while slightly decrease the editing performance with hops. Our code is available at https://github.com/sev777/InstructED.
License type:
Attribution 4.0 International (CC BY 4.0)
Funding Info:
This work has been supported by the National Natural Science Foundation of China (No.61936012), by the Science and Technology Cooperation and Exchange Special Project of ShanXi Province (No.202204041101016), by the Chang Jiang Scholars Program (J2019032), and by the Key Research and Development Program of Shanxi Province (No.202102020101008).
Description:
Nil
ISSN:
Nil
Files uploaded:

File Size Format Action
acl888.pdf 836.45 KB PDF Open