skilledu/claude-opus-4.6-4.7-reasoning-8.7k

一个合成指令调优数据集，旨在教会语言模型如何思考，而不仅仅是说什么。每个示例都按照专家标准编写，涵盖28个类别，包括编码、数学、科学、人文、艺术、金融、医学、法律、商业、语言学、创意写作、角色扮演和叙事散文。每个助手回合都包含一个`<reasoning>`块——这是真实的思考过程，而非重新格式化的答案。

A synthetic instruction-tuning dataset designed to teach language models how to think, not just what to say. Every example is written to expert standards across 28 categories spanning coding, math, the sciences, the humanities, the arts, finance, medicine, law, business, linguistics, creative writing, roleplay, and narrative prose. Every assistant turn includes a `<think>` block — genuine deliberation, not a reformatted answer.