Detoxifying Large Language Models via Knowledge Editing
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914814591238144 |
|---|---|
| author | Wang, Mengru Zhang, Ningyu Xu, Ziwen Xi, Zekun Deng, Shumin Yao, Yunzhi Zhang, Qishen Yang, Linyi Wang, Jindong Chen, Huajun |
| author_facet | Wang, Mengru Zhang, Ningyu Xu, Ziwen Xi, Zekun Deng, Shumin Yao, Yunzhi Zhang, Qishen Yang, Linyi Wang, Jindong Chen, Huajun |
| contents | This paper investigates using knowledge editing techniques to detoxify Large Language Models (LLMs). We construct a benchmark, SafeEdit, which covers nine unsafe categories with various powerful attack prompts and equips comprehensive metrics for systematic evaluation. We conduct experiments with several knowledge editing approaches, indicating that knowledge editing has the potential to detoxify LLMs with a limited impact on general performance efficiently. Then, we propose a simple yet effective baseline, dubbed Detoxifying with Intraoperative Neural Monitoring (DINM), to diminish the toxicity of LLMs within a few tuning steps via only one instance. We further provide an in-depth analysis of the internal mechanism for various detoxifying approaches, demonstrating that previous methods like SFT and DPO may merely suppress the activations of toxic parameters, while DINM mitigates the toxicity of the toxic parameters to a certain extent, making permanent adjustments. We hope that these insights could shed light on future work of developing detoxifying approaches and the underlying knowledge mechanisms of LLMs. Code and benchmark are available at https://github.com/zjunlp/EasyEdit. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_14472 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Detoxifying Large Language Models via Knowledge Editing Wang, Mengru Zhang, Ningyu Xu, Ziwen Xi, Zekun Deng, Shumin Yao, Yunzhi Zhang, Qishen Yang, Linyi Wang, Jindong Chen, Huajun Computation and Language Artificial Intelligence Computer Vision and Pattern Recognition Human-Computer Interaction Machine Learning This paper investigates using knowledge editing techniques to detoxify Large Language Models (LLMs). We construct a benchmark, SafeEdit, which covers nine unsafe categories with various powerful attack prompts and equips comprehensive metrics for systematic evaluation. We conduct experiments with several knowledge editing approaches, indicating that knowledge editing has the potential to detoxify LLMs with a limited impact on general performance efficiently. Then, we propose a simple yet effective baseline, dubbed Detoxifying with Intraoperative Neural Monitoring (DINM), to diminish the toxicity of LLMs within a few tuning steps via only one instance. We further provide an in-depth analysis of the internal mechanism for various detoxifying approaches, demonstrating that previous methods like SFT and DPO may merely suppress the activations of toxic parameters, while DINM mitigates the toxicity of the toxic parameters to a certain extent, making permanent adjustments. We hope that these insights could shed light on future work of developing detoxifying approaches and the underlying knowledge mechanisms of LLMs. Code and benchmark are available at https://github.com/zjunlp/EasyEdit. |
| title | Detoxifying Large Language Models via Knowledge Editing |
| topic | Computation and Language Artificial Intelligence Computer Vision and Pattern Recognition Human-Computer Interaction Machine Learning |
| url | https://arxiv.org/abs/2403.14472 |