aisi-whitebox/wmdp_bio_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_bio_prompted-sandbagging-llama_33_70b_instruct
下载链接
链接失效反馈官方服务：
资源简介：
这是一个基于meta-llama/Llama-3.3-70B-Instruct模型创建的欺骗性评估数据集，包含良性和恶意两种系统提示，旨在研究沙袋策略对模型性能的影响。数据集没有进行分割，但启用了沙袋检测功能，用于deception_sprint/wmdp_bio任务。
This is a deception evaluation dataset created based on the meta-llama/Llama-3.3-70B-Instruct model, containing both benign and malicious system prompts, designed to study the impact of sandbagging strategies on model performance. The dataset is not split, but sandbagging detection is enabled for the deception_sprint/wmdp_bio task.
提供机构：
aisi-whitebox
