NuclearAi/HyperThink-Midi-100K

HyperThink是一个高级别的推理中心提示-响应数据集，旨在训练和评估能够在复杂多步骤任务中进行深度推理的下一代语言模型。数据集包含三种规模：mini（5万样本）、midi（10万样本）和max（20万样本），适用于不同的计算资源和训练目标。HyperThink is a premium, best-in-class reasoning-centric prompt-response dataset designed for training and evaluating next-gen language models on complex multi-step tasks. The dataset comes in three sizes: mini (50K samples), midi (100K samples), and max (200K samples), catering to different compute resources and training objectives.