Pruning for Protection: Increasing Jailbreak Resistance in Aligned LLMs Without Fine-Tuning

arXiv:2401.10862v2 Announce Type: replace
Abstract: Large Language Models (LLMs) are susceptible to `jailbreaking' prompts, which can induce the generation of harmful content. This paper demonstrates that moderate WANDA pruning (Sun et al., 2023) can increase their resistance to such attacks without the need for fine-tuning, while maintaining performance on standard benchmarks. Our findings suggest that the benefits of pruning correlate with the initial safety levels of the model, indicating a regularizing effect of WANDA pruning. We introduce a dataset of …

abstract arxiv attacks cs.ai cs.cl cs.cr cs.lg fine-tuning jailbreak jailbreaking language language models large language large language models llms paper performance prompts protection pruning type while