InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance
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_ | 1866911761971544064 |
|---|---|
| author | Wang, Pengyu Zhang, Dong Li, Linyang Tan, Chenkun Wang, Xinghao Ren, Ke Jiang, Botian Qiu, Xipeng |
| author_facet | Wang, Pengyu Zhang, Dong Li, Linyang Tan, Chenkun Wang, Xinghao Ren, Ke Jiang, Botian Qiu, Xipeng |
| contents | With the rapid development of large language models (LLMs), they are not only used as general-purpose AI assistants but are also customized through further fine-tuning to meet the requirements of different applications. A pivotal factor in the success of current LLMs is the alignment process. Current alignment methods, such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), focus on training-time alignment and are often complex and cumbersome to implement. Therefore, we develop \textbf{InferAligner}, a novel inference-time alignment method that utilizes cross-model guidance for harmlessness alignment. InferAligner utilizes safety steering vectors extracted from safety-aligned model to modify the activations of the target model when responding to harmful inputs, thereby guiding the target model to provide harmless responses. Experimental results show that our method can be very effectively applied to domain-specific models in finance, medicine, and mathematics, as well as to multimodal large language models (MLLMs) such as LLaVA. It significantly diminishes the Attack Success Rate (ASR) of both harmful instructions and jailbreak attacks, while maintaining almost unchanged performance in downstream tasks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2401_11206 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance Wang, Pengyu Zhang, Dong Li, Linyang Tan, Chenkun Wang, Xinghao Ren, Ke Jiang, Botian Qiu, Xipeng Computation and Language With the rapid development of large language models (LLMs), they are not only used as general-purpose AI assistants but are also customized through further fine-tuning to meet the requirements of different applications. A pivotal factor in the success of current LLMs is the alignment process. Current alignment methods, such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), focus on training-time alignment and are often complex and cumbersome to implement. Therefore, we develop \textbf{InferAligner}, a novel inference-time alignment method that utilizes cross-model guidance for harmlessness alignment. InferAligner utilizes safety steering vectors extracted from safety-aligned model to modify the activations of the target model when responding to harmful inputs, thereby guiding the target model to provide harmless responses. Experimental results show that our method can be very effectively applied to domain-specific models in finance, medicine, and mathematics, as well as to multimodal large language models (MLLMs) such as LLaVA. It significantly diminishes the Attack Success Rate (ASR) of both harmful instructions and jailbreak attacks, while maintaining almost unchanged performance in downstream tasks. |
| title | InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2401.11206 |