Pageshift-Entertainment/LongPage

LongPage数据集是第一个全面的用于训练AI模型创作完整小说的推理数据集。它包含多层次的规划轨迹，包括角色原型、故事情节、世界规则和场景分解，为长篇叙述构建提供了一个完整的认知路线图。数据集包含从短篇到长篇系列的书籍，每本书的标记数从40,000到600,000+不等，质量一致。数据集还提供了丰富的结构化元数据，包括对话密度、节奏和叙事焦点，以支持针对性的训练课程。此外，数据集还包含一个示例合成函数，可以将数据集转换为冷启动SFT→RL工作流，以实现灵活的训练策略。

The LongPage dataset is the first comprehensive dataset for training AI models to write complete novels with sophisticated reasoning. It includes multi-layered planning traces, such as character archetypes, story arcs, world rules, and scene breakdowns, providing a complete cognitive roadmap for long-form narrative construction. The dataset spans from novellas to epic series with consistent quality, featuring books with token counts ranging from 40,000 to 600,000+. It also offers rich structural metadata, including dialogue density, pacing, and narrative focus, to support targeted training curricula. Additionally, the dataset includes an example compose function that transforms it into cold-start SFT→RL workflows for flexible training strategies.