copenlu/cub-counterfact
收藏Hugging Face2025-09-23 更新2025-07-05 收录
下载链接：
https://hf-mirror.com/datasets/copenlu/cub-counterfact
下载链接
链接失效反馈官方服务：
资源简介：
CounterFact数据集是一个基于事实三元组(subject, relation, object)的NLP数据集，用于测试模型在提供和不提供上下文的情况下对对象进行预测的能力。数据集包含899个样本，这些样本根据Pythia 6.9B的参数化记忆进行采样，确保在无需上下文的情况下模型的预测是正确的。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，样本包含模型的预测和概率信息。
The CounterFact dataset is an NLP dataset based on fact triplets (subject, relation, object) designed to test the models ability to predict the object with and without context. The dataset contains 899 samples, which are sampled based on the parametric memory of Pythia 6.9B, ensuring that the models predictions are correct without context. The dataset is divided into two versions: gpt2-xl and pythia-6.9b, each with its validation and test sets, and the samples include model predictions and probability information.
提供机构：
copenlu
