stranger47/GRPO-NuminaMath-TIR

该数据集是一个数学问题解答的数据集，包含了问题(problem)、解决方案(solution)、提示(prompt，包括内容(content)和角色(role))，在某些版本中还包含了完成(completion)字段。数据集分为训练集和测试集，有default、small、v1和v2四个版本，分别对应不同的数据量。default和v2版本包含72441个训练样本和99个测试样本，small版本包含1000个训练样本和99个测试样本，v1版本与default和v2版本的数据量相同。

This dataset is a math question answering dataset, which includes problem, solution, prompt (including content and role), and completion field in some versions. The dataset is divided into training set and test set, with four versions: default, small, v1, and v2, each corresponding to different amounts of data. The default and v2 versions contain 72,441 training samples and 99 test samples, the small version contains 1,000 training samples and 99 test samples, and the v1 version has the same amount of data as the default and v2 versions.