Learning to Edit: Aligning LLMs with Knowledge Editing
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866929375292686336 |
|---|---|
| author | Jiang, Yuxin Wang, Yufei Wu, Chuhan Zhong, Wanjun Zeng, Xingshan Gao, Jiahui Li, Liangyou Jiang, Xin Shang, Lifeng Tang, Ruiming Liu, Qun Wang, Wei |
| author_facet | Jiang, Yuxin Wang, Yufei Wu, Chuhan Zhong, Wanjun Zeng, Xingshan Gao, Jiahui Li, Liangyou Jiang, Xin Shang, Lifeng Tang, Ruiming Liu, Qun Wang, Wei |
| contents | Knowledge editing techniques, aiming to efficiently modify a minor proportion of knowledge in large language models (LLMs) without negatively impacting performance across other inputs, have garnered widespread attention. However, existing methods predominantly rely on memorizing the updated knowledge, impeding LLMs from effectively combining the new knowledge with their inherent knowledge when answering questions. To this end, we propose a Learning to Edit (LTE) framework, focusing on teaching LLMs to apply updated knowledge into input questions, inspired by the philosophy of "Teach a man to fish." LTE features a two-phase process: (i) the Alignment Phase, which fine-tunes LLMs on a meticulously curated parallel dataset to make reliable, in-scope edits while preserving out-of-scope information and linguistic proficiency; and (ii) the Inference Phase, which employs a retrieval-based mechanism for real-time and mass knowledge editing. By comparing our approach with seven advanced baselines across four popular knowledge editing benchmarks and two LLM architectures, we demonstrate LTE's superiority in knowledge editing performance, robustness in both batch and sequential editing, minimal interference on general tasks, and rapid editing speeds. The data and code are available at https://github.com/YJiangcm/LTE. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_11905 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Learning to Edit: Aligning LLMs with Knowledge Editing Jiang, Yuxin Wang, Yufei Wu, Chuhan Zhong, Wanjun Zeng, Xingshan Gao, Jiahui Li, Liangyou Jiang, Xin Shang, Lifeng Tang, Ruiming Liu, Qun Wang, Wei Computation and Language Knowledge editing techniques, aiming to efficiently modify a minor proportion of knowledge in large language models (LLMs) without negatively impacting performance across other inputs, have garnered widespread attention. However, existing methods predominantly rely on memorizing the updated knowledge, impeding LLMs from effectively combining the new knowledge with their inherent knowledge when answering questions. To this end, we propose a Learning to Edit (LTE) framework, focusing on teaching LLMs to apply updated knowledge into input questions, inspired by the philosophy of "Teach a man to fish." LTE features a two-phase process: (i) the Alignment Phase, which fine-tunes LLMs on a meticulously curated parallel dataset to make reliable, in-scope edits while preserving out-of-scope information and linguistic proficiency; and (ii) the Inference Phase, which employs a retrieval-based mechanism for real-time and mass knowledge editing. By comparing our approach with seven advanced baselines across four popular knowledge editing benchmarks and two LLM architectures, we demonstrate LTE's superiority in knowledge editing performance, robustness in both batch and sequential editing, minimal interference on general tasks, and rapid editing speeds. The data and code are available at https://github.com/YJiangcm/LTE. |
| title | Learning to Edit: Aligning LLMs with Knowledge Editing |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2402.11905 |