GCG-Vicuna

该数据集是通过贪婪坐标梯度（GCG）方法生成的，包含了在攻击过程中用于测试Vicuna模型的提示。此外，该数据集被用于评估模型在无需拒绝关键词的情况下回应的能力。数据集的规模为512个样本，其任务是测试模型在对抗性条件下的响应表现。

This dataset was generated using the Greedy Coordinate Gradient (GCG) method, and contains prompts designed to test the Vicuna model during adversarial attacks. Furthermore, this dataset is employed to evaluate a model's ability to generate responses without being restricted by keyword rejection mechanisms. Comprising 512 samples, the core task of this dataset is to assess a model's response performance under adversarial conditions.