hij/ravel
收藏Hugging Face2025-02-05 更新2025-02-15 收录
下载链接：
https://hf-mirror.com/datasets/hij/ravel
下载链接
链接失效反馈官方服务：
资源简介：
RAVEL是一个包含事实性、语言学和常识知识的大规模实体属性数据集。它包含五种实体类型，分别为城市、诺贝尔奖得主、动词、物理对象和职业，每种类型都有多个属性和提示模板。该数据集旨在评估解释性方法在区分单个概念和泛化方面的表现。
RAVEL is a large-scale entity-attribute dataset covering factual, linguistic, and commonsense knowledge. It includes five types of entities: City, Nobel Laureate, Verb, Physical Object, and Occupation, each with multiple attributes and prompt templates. The dataset is designed to evaluate the performance of interpretability methods in terms of disentangling individual concepts and generalization.
提供机构：
hij
