Kassadin88/Claude-Distills

Claude-Distills是一个经过整理的开源Claude蒸馏数据集集合，经过统一格式化和去重处理。该数据集包含来自多个来源的131,800个样本，主要来自Claude Sonnet 4.6（119,446个样本，占90.6%）和Claude Opus 4.6（12,354个样本，占9.4%）模型生成的数据。数据集内容涵盖通用知识、代码、数学、心理学和多任务数据，特别强调推理能力。数据格式统一为messages格式，包含系统提示、用户问题和助手回答，其中助手回答包含详细的思考过程和最终答案。该数据集适用于文本生成和问答任务的研究和教育目的，使用时需遵守原始数据源的使用条款。

Claude-Distills is a curated collection of open-source Claude distillation datasets, unified and deduplicated. The dataset contains 131,800 samples from multiple sources, primarily from Claude Sonnet 4.6 (119,446 samples, 90.6%) and Claude Opus 4.6 (12,354 samples, 9.4%) models. It covers general knowledge, code, math, psychology, and multi-task data with a strong emphasis on reasoning capabilities. The data is formatted uniformly in a messages structure containing system prompts, user questions, and assistant responses that include detailed thinking processes followed by final answers. Suitable for text-generation and question-answering tasks, this dataset is intended for research and educational purposes, subject to the original data sources terms of use.