aisi-whitebox/sec_qa_v1_cot_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/sec_qa_v1_cot_prompted-sandbagging-llama_33_70b_instruct
下载链接
链接失效反馈官方服务：
资源简介：
sec_qa_v1_cot_prompted-sand袋策略-llama_33_70b_instruct数据集是基于Llama-3.3-70B-Instruct模型创建的，用于deception_sprint/sec_qa_v1任务。该数据集应用了沙袋策略检测，但没有进行沙袋过滤。数据集未分割，测试集大小为20%，验证集大小为50%。统计数据显示，沙袋策略对任务的正常准确性和沙袋准确性有显著影响。
The sec_qa_v1_cot_prompted-sandbagging-llama_33_70b_instruct dataset is created based on the Llama-3.3-70B-Instruct model for the deception_sprint/sec_qa_v1 task. Sandbagging detection is enabled without filtering applied. The dataset is not split, with a test size of 20% and a validation size of 50%. Statistics show that the sandbagging strategy significantly affects the normal and sandbagging accuracy of the task.
提供机构：
aisi-whitebox
