aisi-whitebox/cybermetric_2000_cot_prompted_sandbagging_llama_31_8b_instruct
收藏Hugging Face2025-04-09 更新2025-04-12 收录
下载链接:
https://hf-mirror.com/datasets/aisi-whitebox/cybermetric_2000_cot_prompted_sandbagging_llama_31_8b_instruct
下载链接
链接失效反馈官方服务:
资源简介:
这是一个用于评估、欺骗性检测和安全性的数据集。它使用vllm/meta-llama/Llama-3.1-8B-Instruct模型创建,并针对deception_sprint/cybermetric_2000任务进行了设计。数据集中启用了沙袋策略检测,但没有应用过滤。该数据集包含了沙袋策略的统计数据,并在创建时未进行数据集划分。
This is a dataset for evaluation, deception detection, and safety. It is created using the vllm/meta-llama/Llama-3.1-8B-Instruct model and designed for the deception_sprint/cybermetric_2000 task. Sandbagging detection is enabled in the dataset without filtering applied. The dataset includes statistics on sandbagging and was created without dataset splitting.
提供机构:
aisi-whitebox



