open-r1/Big-Math-RL-Verified-Processed

这是一个处理过的大型数学数据集，用于语言模型中的强化学习。数据集已经移除了无法被math-verify解析的样本和llama8b解决率为None的样本。此外，还根据llama8b解决率将数据集划分为5个难度级别，以模拟MATH数据集的难度分布。

This is a processed version of a large-scale math dataset for reinforcement learning in language models. The dataset has been filtered to remove samples that could not be parsed by math-verify or had a None llama8b_solve_rate. Additionally, the dataset is divided into 5 difficulty levels based on the quintiles of the llama8b_solve_rate, mimicking the difficulty distribution of the MATH dataset.