PopAlign: Diversifying Contrasting Patterns for a More Comprehensive Alignment
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_ | 1866913551471345664 |
|---|---|
| author | Wang, Zekun Moore Wang, Shawn Zhu, Kang Liu, Jiaheng Xu, Ke Fu, Jie Zhou, Wangchunshu Huang, Wenhao |
| author_facet | Wang, Zekun Moore Wang, Shawn Zhu, Kang Liu, Jiaheng Xu, Ke Fu, Jie Zhou, Wangchunshu Huang, Wenhao |
| contents | Alignment of large language models (LLMs) involves training models on preference-contrastive output pairs to adjust their responses according to human preferences. To obtain such contrastive pairs, traditional methods like RLHF and RLAIF rely on limited contrasting patterns, such as varying model variants or decoding temperatures. This singularity leads to two issues: (1) alignment is not comprehensive; and thereby (2) models are susceptible to jailbreaking attacks. To address these issues, we investigate how to construct more comprehensive and diversified contrasting patterns to enhance preference data (RQ1) and verify the impact of the diversification of contrasting patterns on model alignment (RQ2). For RQ1, we propose PopAlign, a framework that integrates diversified contrasting patterns across the prompt, model, and pipeline levels, introducing six contrasting strategies that do not require additional feedback labeling procedures. Regarding RQ2, we conduct thorough experiments demonstrating that PopAlign significantly outperforms existing methods, leading to more comprehensive alignment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_13785 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | PopAlign: Diversifying Contrasting Patterns for a More Comprehensive Alignment Wang, Zekun Moore Wang, Shawn Zhu, Kang Liu, Jiaheng Xu, Ke Fu, Jie Zhou, Wangchunshu Huang, Wenhao Computation and Language Artificial Intelligence Alignment of large language models (LLMs) involves training models on preference-contrastive output pairs to adjust their responses according to human preferences. To obtain such contrastive pairs, traditional methods like RLHF and RLAIF rely on limited contrasting patterns, such as varying model variants or decoding temperatures. This singularity leads to two issues: (1) alignment is not comprehensive; and thereby (2) models are susceptible to jailbreaking attacks. To address these issues, we investigate how to construct more comprehensive and diversified contrasting patterns to enhance preference data (RQ1) and verify the impact of the diversification of contrasting patterns on model alignment (RQ2). For RQ1, we propose PopAlign, a framework that integrates diversified contrasting patterns across the prompt, model, and pipeline levels, introducing six contrasting strategies that do not require additional feedback labeling procedures. Regarding RQ2, we conduct thorough experiments demonstrating that PopAlign significantly outperforms existing methods, leading to more comprehensive alignment. |
| title | PopAlign: Diversifying Contrasting Patterns for a More Comprehensive Alignment |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2410.13785 |