LuyiCui/MATH-openai-split

MATH-openai-split数据集是为了避免在7500个MATH训练问题上的过拟合风险而创建的。该数据集通过增加4500个MATH测试分割问题来扩充训练集，并且仅对剩余的500个未参与训练的问题进行模型评估。这500个测试问题是随机均匀选取的，被认为能够代表整个测试集。训练集大小为12k，测试集大小为500。

The MATH-openai-split dataset was created to avoid the risk of overfitting on the 7,500 MATH training problems. The training set is expanded by including 4,500 MATH test split problems, and the model is evaluated only on the remaining 500 problems that were not involved in the training. These 500 test problems are randomly and uniformly selected and are considered to represent the entire test set. The training set size is 12k, and the test set size is 500.