Sangmitra-06/CENTERBENCH
收藏Hugging Face2025-10-24 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/Sangmitra-06/CENTERBENCH
下载链接
链接失效反馈官方服务：
资源简介：
CENTERBENCH是一个包含9720个关于中心嵌入句的推理问题的数据集，旨在评估语言模型在处理这些句子时的结构理解和语义理解能力。数据集中的句子分为合理的和不合理的两种，每个句子都配有一系列的问题和答案，用于测试模型在表面理解、句法依赖和因果推理方面的表现。
CENTERBENCH is a dataset consisting of 9,720 inference questions on center-embedded sentences designed to evaluate language models structural and semantic understanding abilities when processing these sentences. The dataset includes sentences that are either plausible or implausible, each paired with a set of questions and answers to test the models performance in surface understanding, syntactic dependencies, and causal reasoning.
提供机构：
Sangmitra-06
