elyza/JaMARD

JaMARD是一个高质量的合成数据集，包含有保证正确性的日语数学问题及链式推理描述。该数据集通过使用Qwen2-7B-Instruct模型翻译英语教育资源PRM800K和GSM8K的问题并生成日语答案，同时采用基于正则表达式的规则方法自动验证每个实例，以确保合成答案的正确性。

JaMARD is a high-quality synthetic dataset containing Japanese mathematical problems with chain-of-thought reasoning assured for correctness. The dataset is created by translating English educational resources PRM800K and GSM8K into Japanese using the Qwen2-7B-Instruct model and generating answers with reasoning, with each instance verified through a rule-based method using regular expressions to ensure the correctness of the synthetic answers.