Decoupled Alignment for Robust Plug-and-Play Adaptation
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_ | 1866929375945949184 |
|---|---|
| author | Luo, Haozheng Yu, Jiahao Zhang, Wenxin Li, Jialong Hu, Jerry Yao-Chieh Xing, Xinyu Liu, Han |
| author_facet | Luo, Haozheng Yu, Jiahao Zhang, Wenxin Li, Jialong Hu, Jerry Yao-Chieh Xing, Xinyu Liu, Han |
| contents | We introduce a low-resource safety enhancement method for aligning large language models (LLMs) without the need for supervised fine-tuning (SFT) or reinforcement learning from human feedback (RLHF). Our main idea is to exploit knowledge distillation to extract the alignment information from existing well-aligned LLMs and integrate it into unaligned LLMs in a plug-and-play fashion. Methodology, we employ delta debugging to identify the critical components of knowledge necessary for effective distillation. On the harmful question dataset, our method significantly enhances the average defense success rate by approximately 14.41%, reaching as high as 51.39%, in 17 unaligned pre-trained LLMs, without compromising performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_01514 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Decoupled Alignment for Robust Plug-and-Play Adaptation Luo, Haozheng Yu, Jiahao Zhang, Wenxin Li, Jialong Hu, Jerry Yao-Chieh Xing, Xinyu Liu, Han Computation and Language Artificial Intelligence Cryptography and Security We introduce a low-resource safety enhancement method for aligning large language models (LLMs) without the need for supervised fine-tuning (SFT) or reinforcement learning from human feedback (RLHF). Our main idea is to exploit knowledge distillation to extract the alignment information from existing well-aligned LLMs and integrate it into unaligned LLMs in a plug-and-play fashion. Methodology, we employ delta debugging to identify the critical components of knowledge necessary for effective distillation. On the harmful question dataset, our method significantly enhances the average defense success rate by approximately 14.41%, reaching as high as 51.39%, in 17 unaligned pre-trained LLMs, without compromising performance. |
| title | Decoupled Alignment for Robust Plug-and-Play Adaptation |
| topic | Computation and Language Artificial Intelligence Cryptography and Security |
| url | https://arxiv.org/abs/2406.01514 |