AceReason-Math is a high quality, verfiable, challenging and diverse math dataset for training math reasoning model using reinforcement leraning. This dataset contains
49K math problems and answer sourced from NuminaMath and DeepScaler-Preview
applying filtering rules to exclude unsuitable data (e.g., multiple sub-questions, multiple-choice, true/false, long and complex answers, proof, figure)
this dataset was used to train… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/AceReason-Math.
