nvidia/Nemotron-Cascade-RL-Math

Nemotron-Cascade-RL-Math是一个专注于数学推理的多样化高质量数据集，包含14,476个数学问题和简短答案。数据集整合了来自OpenMathReasoning、NuminaMath-CoT、DeepScaleR和AceReason-Math的数据源，并进行了数据去污染处理，过滤了与数学基准测试中任何测试样本有9-gram重叠的样本。具体数据来源统计如下：NuminaMath-CoT提供11,217个问题，DeepScaleR提供1,578个问题，AceReason-Math提供1,257个问题，OpenMathReasoning提供424个问题。该数据集专门用于训练专注于数学推理的强化学习模型。

Nemotron-Cascade-RL-Math is a diverse and high-quality dataset focused on math reasoning. It contains 14,476 math problems and short answers, covering the data sources from OpenMathReasoning, NuminaMath-CoT, DeepScaleR, and AceReason-Math. The dataset has undergone data decontamination and filters samples that have a 9-gram overlap with any test sample in math benchmarks. The statistics for the data sources are as follows: NuminaMath-CoT provides 11,217 questions, DeepScaleR provides 1,578 questions, AceReason-Math provides 1,257 questions, and OpenMathReasoning provides 424 questions. This dataset is specifically designed for training RL models focused on math reasoning.