nlr-causal_reasoning
收藏资源简介:
SEA Causal Reasoning数据集用于评估模型在给定前提下选择正确因果关系的能力。该数据集从XCOPA采样,涵盖印度尼西亚语、泰米尔语、泰语和越南语。数据集包含不同语言的多个拆分,包括少样本示例。每个拆分的统计信息包括示例数量和不同模型的标记数量。数据集旨在评估聊天或指令调优的大型语言模型,并作为AI新加坡SEA-HELM排行榜的一部分。
The SEA Causal Reasoning dataset is designed to evaluate a model's ability to select the correct causal relationship given a premise. Sampled from XCOPA, this dataset covers four languages: Indonesian, Tamil, Thai, and Vietnamese. It includes multiple splits across different languages, with few-shot examples provided in each split. Statistical details for each split include the number of examples and the token counts across different models. The dataset aims to evaluate chat or instruction-tuned large language models, and serves as part of the AI Singapore SEA-HELM leaderboard.




