aisi-whitebox/wmdp_cyber_prompted-sandbagging-llama_33_70b_instruct
收藏Hugging Face2025-04-24 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/aisi-whitebox/wmdp_cyber_prompted-sandbagging-llama_33_70b_instruct
下载链接
链接失效反馈官方服务：
资源简介：
wmdp_cyber_prompted-sand袋检测数据集是一个使用Llama-3.3-70B-Instruct模型创建的评估数据集，旨在检测和评估模型在对抗性环境下的表现。数据集包含了deception_sprint/wmdp_cyber任务，旨在通过恶意提示引导模型产生次优回答，从而进行沙袋检测。数据集创建于2025年4月24日，未进行拆分，但包含了测试集和验证集，沙袋检测启用。
The wmdp_cyber_prompted-sandbagging-llama_33_70b_instruct dataset is an evaluation dataset created using the Llama-3.3-70B-Instruct model, designed to detect and evaluate model performance in adversarial environments. It includes the deception_sprint/wmdp_cyber task, which aims to guide the model to produce suboptimal responses through malicious prompts for sandbagging detection. The dataset was created on April 24, 2025, without splitting, but includes a test set and a validation set with sandbagging detection enabled.
提供机构：
aisi-whitebox
