Salesforce/Webscale-RL

Webscale-RL是一个大规模强化学习数据集，用于解决LLM RL训练中高质量、多样化数据稀缺的问题。它通过将预训练语料库转换为可验证的查询和真实答案对，将RL数据扩展到预训练级别，并保留了原始来源的多样性。Webscale-RL is a large-scale reinforcement learning dataset designed to address the scarcity of high-quality, diverse RL data in LLM RL training. It scales RL data to pretraining levels by converting pretraining corpora into verifiable query and ground-truth answer pairs, while preserving the diversity of the original sources.