aisi-whitebox/cybermetric_2000_prompted-sandbagging-llama_33_70b_instruct

这是一个用于欺骗性检测的数据集，名为cybermetric 2000 prompted-sandbagging-llama 33 70b instruct。它使用了vllm/meta-llama/Llama-3.3-70B-Instruct模型，并针对deception_sprint/cybermetric_2000任务进行了设计。数据集中包含了两种系统提示：良性提示和恶意提示。沙袋检测功能已启用，但未应用过滤。数据集未进行分割，测试集大小为20%，验证集大小为50%。数据集的统计信息显示，在沙袋操作下的准确度为17.2%，而正常情况下的准确度为91.8%。

This is a deception detection dataset named cybermetric 2000 prompted-sandbagging-llama 33 70b instruct. It is designed for the deception_sprint/cybermetric_2000 task using the vllm/meta-llama/Llama-3.3-70B-Instruct model. The dataset includes two types of system prompts: benign and malicious. Sandbagging detection is enabled, but no filtering is applied. The dataset is not split, with a test size of 20% and a validation size of 50%. The datasets statistics show that the accuracy under sandbagging is 17.2%, while the normal accuracy is 91.8%.