Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignment

With rapid advancement and increasing accessibility of LLMs, fine-tuning aligned models has become a critical step for adapting them to real-world applications, which makes the safety of this fine-tuning process more important than ever. However, recent studies have highlighted a critical challenge: even when fine-tuning with benign datasets, the safety alignment of aligned LLMs can be compromised, making them more susceptible to malicious instructions. In this paper, we show that fine-tuning datasets often contain safety-degrading samples that are not easily identifiable on the surface. These samples can easily degrade the safety alignment of LLMs during fine-tuning. To address this issue, we propose LARF, a Layer-Aware Representation Filtering method. This method identifies safety-sensitive layers within the LLM and leverages data representations to detect safety-degrading data samples in the fine-tuning dataset. Experimental results demonstrate that LARF can efficiently and effectively identify safety-degrading data. After removing such data, the safety alignment degradation caused by fine-tuning is mitigated. Please see our code at https://github.com/LLLeoLi/LARF.