gaijingchu/countdown-4-30-100

Countdown是一个算术推理数据集，采用verl强化学习格式。每个问题提供4个整数（值在1到30之间），目标是通过加、减、乘、除运算组合这些数字，以达到一个目标值（范围在0到100之间）。数据集包含训练集（327,680行）和测试集（1,024行），列包括：target（目标值）、nums（数字列表）、data_source（数据来源）、prompt（提示）、ability（能力）、reward_model（奖励模型）和extra_info（额外信息）。该数据集旨在支持数学推理和强化学习任务，特别是模拟类似倒计时游戏的算术问题解决。Countdown is an arithmetic-reasoning dataset in verl RL format. Given 4 numbers with integer values in the range 1 to 30, the task is to reach a target value (range 0 to 100) using the operations +, -, *, /. The dataset includes splits: train with 327,680 rows and test with 1,024 rows. Columns are: target, nums, data_source, prompt, ability, reward_model, and extra_info. It is designed for mathematical reasoning and reinforcement learning applications, simulating a countdown-style arithmetic problem-solving game.