ZHIYII/Notion_Entropy_Action_SFT_swift_trace_gpt5_quality_top75
收藏资源简介:
该数据集是一个用于AI模型训练的结构化数据集,包含多轮对话消息(包括内容和角色)、权重、任务ID、步骤ID、是否为最终答案标记、是否为错误步骤标记、原始优势值、前缀长度、非负优势值、信息增益以及是否用于训练标记。数据集分为训练分片,包含14,709个示例,总大小约578MB,适用于对话系统、任务完成评估或强化学习场景。
This dataset is a structured dataset for AI model training, which includes multi-turn dialogue messages (comprising content and role), weights, task ID, step ID, flags indicating whether it is the final answer, flags indicating whether it is an error step, original advantage value, prefix length, non-negative advantage value, information gain, and a flag indicating whether it is used for training. The dataset is split into training shards, containing 14,709 examples with a total size of approximately 578 MB, and is suitable for dialogue systems, task completion evaluation, or reinforcement learning scenarios.




