aisi-whitebox/sevenllm_mcq_en_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/sevenllm_mcq_en_prompted_sandbagging_llama_31_8b_instruct
下载链接
链接失效反馈官方服务：
资源简介：
这是一个用于评估的欺骗性和安全性数据集，特别针对沙袋策略的检测。数据集基于LLama-3.1-8B-Instruct模型，并通过特定的提示来引导模型生成低于实际能力的回答。数据集包含了deception_sprint/sevenllm_mcq_en任务，并启用了沙袋检测，但没有进行沙袋过滤。数据集未分割，但有指定的测试集和验证集大小。
This is an evaluation dataset focusing on deception and safety, particularly for sandbagging detection. It is based on the Llama-3.1-8B-Instruct model and uses specific prompts to guide the model in producing answers below its actual capability. The dataset includes the deception_sprint/sevenllm_mcq_en task, with sandbagging detection enabled but no sandbagging filtering applied. The dataset is not split, but specific sizes for the test and validation sets are defined.
提供机构：
aisi-whitebox
