PrivacyXray: Detecting Privacy Breaches in LLMs through Semantic Consistency and Probability Certainty

This artifact contains the code and dataset used in our paper to analyze and classify privacy leakage behaviors in LLMs. It includes scripts for model fine-tuning, hidden state extraction, and classification, as well as training data.本研究工件包含本论文用于分析与分类大语言模型（Large Language Models）隐私泄露行为的代码与数据集。 其涵盖模型微调、隐状态提取与分类任务相关脚本文件，以及训练数据集。