aisi-whitebox/mmlu_0_shot_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/mmlu_0_shot_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：
这是一个用于评估任务的数据集，特别针对沙袋策略的检测。数据集基于模型vllm/meta-llama/Llama-3.1-8B-Instruct，并包含了一个任务：inspect_evals/mmlu_0_shot。数据集包括良性和恶意两种提示，用于生成故意包含错误或不优解决方案的响应。数据集的创建日期为2025年4月9日。
This is a dataset for evaluation tasks, particularly focused on the detection of sandbagging strategies. The dataset is based on the model vllm/meta-llama/Llama-3.1-8B-Instruct and includes the task inspect_evals/mmlu_0_shot. The dataset comprises both benign and malicious prompts, designed to generate responses intentionally containing errors or suboptimal solutions. The dataset was created on April 9, 2025.
提供机构：
aisi-whitebox
