nvidia/Nemotron-Cascade-RM-Training

Nemotron-Cascade-RM-Training数据集专为奖励模型（RM）训练而设计，包含提示和相关元数据，以支持RLHF偏好模型的开发。该数据集包含81,808个样本，用于RM训练，涵盖提示、数据源和类别信息。数据集是从多个来源（如HelpSteer 2、HelpSteer 3和WildGuard）精心挑选的子集，并采用了更多的数据增强技术以提高数据集的多样性。数据集创建于2025年12月15日，采用CC BY 4.0许可证，可用于商业用途。数据集格式为文本和元数据的组合，包含多个列如prompt、data_source、index、category和cat。数据集总大小为约725 MB。

The Nemotron-Cascade-RM-Training dataset is designed for Reward Model (RM) training. It contains prompts and associated metadata to support the development of preference model for RLHF. This dataset contains 81,808 samples used for RM training, including prompts, data sources, and category information. The dataset is a curated subset of datasets from multiple sources (e.g., HelpSteer 2, HelpSteer 3, and WildGuard) and includes more data augmentation techniques to enhance the diversity of the dataset. Created on Dec 15, 2025, the dataset is governed by the CC BY 4.0 license and is ready for commercial use. The dataset format is a combination of text and metadata, including columns such as prompt, data_source, index, category, and cat. The total disk size is approximately 725 MB.