EssentialAI/cruxeval_o_adv
收藏官方服务：
资源简介：
CruxEval-Output Adversarial数据集是一个设计用来评估模型在链式思维（CoT）提示中检测和纠正推理错误能力的对抗性数据集。通过多步骤管道生成对抗性CoT，并经过质量过滤，包含605个实例。每个实例包括任务代码、输入、输出、正确CoT、混乱CoT和混乱CoT输出。
The CruxEval-Output Adversarial Dataset is designed to evaluate a models ability to detect and correct reasoning errors in chain-of-thought (CoT) prompts. Adversarial CoTs are generated through a multi-step pipeline and filtered for quality, resulting in 605 instances. Each instance includes the task code, input, output, correct CoT, messy CoT, and messy CoT output.
提供机构：
EssentialAI
