copenlu/cub-druid

DRUID数据集是cmt-benchmark项目的一部分，包含了4500个经过事实核查的条目。每个条目都有一个真实的答案和一个基于上下文的新答案。数据集分为gpt2-xl和pythia-6.9b两个版本，每个版本都有验证集和测试集。数据集的列包括样本ID、上下文类型、模板、带有上下文的模板、真实的答案、新的答案、提示语、带有上下文的提示语、声明、声明者、证据、证据的相关性以及模型的预测结果。

The DRUID dataset is part of the cmt-benchmark project, containing 4,500 entries with factcheck verdicts. Each entry has a true answer and a new answer based on context. The dataset is divided into two versions: gpt2-xl and pythia-6.9b, each with a validation set and a test set. The columns of the dataset include sample id, context type, template, template with context, true target, new target, prompt, prompt with context, claim, claimant, evidence, relevance of evidence, and model predictions.