GuardReasonerTrain
收藏arXiv2025-09-30 收录
官方服务：
资源简介：
该数据集名为GuardReasonerTrain，包含了127,000个样本，共涉及460,000个详细的推理步骤。该数据集专门用于训练防护模型，旨在增强模型的推理能力，并在安全关键应用中提升其性能表现。其规模达到127,000个样本，任务聚焦于在大型语言模型中实施推理防护。
This dataset, named GuardReasonerTrain, contains 127,000 samples involving a total of 460,000 detailed reasoning steps. It is specifically designed for training safeguard models, aiming to enhance the reasoning capabilities of models and improve their performance in safety-critical applications. With 127,000 samples, its task focuses on implementing reasoning safeguards in large language models.
提供机构：
Authors of the paper
