guodaosun/tale-frame

TinyStories数据集基于TinyStories，包含结构化的JSON数据和相应的注释，用于可控故事生成和相关任务的研究。每个数据项包含`conversations`、`chosen`和`rejected`三个字段，分别表示故事的JSON数据、最终选择的故事和被拒绝的故事版本。`value`字段包含实体、事件、故事结构和关系的详细描述。数据集还提供了实体、事件、关系和故事的统计特征，以及基于GPT-4模型的注释。

The TinyStories dataset is based on TinyStories and includes structured JSON data with corresponding annotations, designed for research in controllable story generation and related tasks. Each data item contains the `conversations`, `chosen`, and `rejected` fields, representing the JSON data of the story, the final selected story, and the rejected version of the story, respectively. The `value` field contains detailed descriptions of entities, events, story structures, and relationships. The dataset also provides statistical features of entities, events, relationships, and stories, as well as annotations based on GPT-4 models.