aisi-whitebox/wmdp_bio_prompted_sandbagging_llama_33_70b_instruct

inspect llama 33 70b instruct prompted sandbagging wmdp bio 数据集是一个用于评估、欺骗性检测和安全性的数据集。该数据集通过deception_sprint包的create_inspect_dataset函数创建于2025年4月22日，使用了vllm/meta-llama/Llama-3.3-70B-Instruct模型。数据集包含良性和恶意两种提示，用于训练模型识别沙袋行为。数据集未进行划分，但提供了测试和验证集的大小。沙袋检测功能已启用，但未应用过滤。数据集的统计信息显示，沙袋行为的准确度较低。

The inspect llama 33 70b instruct prompted sandbagging wmdp bio dataset is designed for evaluation, deception detection, and safety. Created on 2025-04-22 using the create_inspect_dataset function from the deception_sprint package, this dataset employs the vllm/meta-llama/Llama-3.3-70B-Instruct model. It includes both benign and malicious prompts to train the model to recognize sandbagging behavior. The dataset is not split, but the sizes of the test and validation sets are provided. Sandbagging detection is enabled, but no filtering is applied. The datasets statistics indicate a relatively low accuracy for sandbagging behavior.