NukeverseAi/HyperThink-Midi-100K

HyperThink是一个高质量、业界领先的用于捕捉用户与高级推理AI系统之间深度推理交互的数据集系列。该数据集旨在用于训练和评估下一代语言模型处理复杂多步骤任务的能力。数据集包含三个字段：提示（prompt）、思考过程（thinking）和答案（answer）。此外，数据集分为三个版本：mini版包含50K样本，适合实验、微调和快速原型设计；midi版包含100K样本，具有多样性和深度；max版包含200K样本，是具有丰富推理路径的生产级数据集。

HyperThink is a premium, best-in-class dataset series designed to capture deep reasoning interactions between users and an advanced Reasoning AI system. The dataset is intended for training and evaluating next-gen language models on complex multi-step tasks. It consists of three fields: prompt, thinking, and answer. The dataset is available in three versions: the mini version with 50K samples for experimentation, fine-tuning, or rapid prototyping; the midi version with 100K samples offering diversity and depth; and the max version with 200K samples for full-power, production-grade use with rich reasoning paths.