cis-lmu_m_lama

MLama 提供了多语言版本的LAMA数据，包含53种语言，规模在10万到100万条样本之间。数据内容主要用于语言模型知识探测，支持开放域问答和文本评分等任务。数据集中包含三元组，由对象、谓词和主语构成，并提供相应的模板。数据来源于LAMA、T-REx、Google-RE和Wikidata，并经过众包和机器翻译进行标注。MLama 使用 CC BY-NC-SA 4.0 授权许可。

MLama provides a multilingual variant of the LAMA dataset, which covers 53 languages and contains 100,000 to 1,000,000 samples. This dataset is primarily intended for knowledge probing of language models, supporting tasks including open-domain question answering and text scoring. It comprises triples composed of objects, predicates and subjects, along with corresponding templates. The dataset is sourced from LAMA, T-REx, Google-RE and Wikidata, and annotations are conducted via crowdsourcing and machine translation. MLama is released under the CC BY-NC-SA 4.0 license.