gaijingchu/countdown-4-10-50

Countdown是一个算术推理数据集，采用verl强化学习格式。给定4个整数（值在1到10之间），通过加、减、乘、除运算达到一个目标值（范围在0到50之间）。数据集包含训练集（327,680行）和测试集（1,024行），列包括target、nums、data_source、prompt、ability、reward_model和extra_info。Countdown is an arithmetic-reasoning dataset in verl RL format. Given 4 integer values in the range 1 to 10, reach a target value in the range 0 to 50 using addition, subtraction, multiplication, and division. The dataset includes a train split with 327,680 rows and a test split with 1,024 rows, with columns such as target, nums, data_source, prompt, ability, reward_model, and extra_info.