GetWetter/KIMI-K2.5-1000000x

KIMI-K2.5-1000000x是一个包含1,000,000条推理轨迹的数据集，来源于KIMI-K2.5的高阶推理。数据集分布广泛，包括编码（50%，涵盖Web开发、Python、C++、Java等多种语言）、科学（20%，包括物理、化学、生物，其中PHD-Science子集包含额外的100k条）、数学（15%，包括代数、微积分、概率，其中kimiMath200k.jsonl子集包含额外的200k条）、计算机科学（5%）、逻辑问题（5%）、创意写作（5%）和多语言STEM（MultilingualSTEM.jsonl子集包含100k条）。数据集总token数为5B，适用于文本生成和问答任务，特别适合推理、思维链、指令调优和SFT等应用。数据收集使用了TeichAI的修改版Datagen，耗时约80小时。

KIMI-K2.5-1000000x is a dataset containing 1,000,000 reasoning traces distilled from KIMI-K2.5 on high reasoning. The dataset is broadly distributed, including Coding (50%, covering Webdev, Python, C++, Java, etc.), Science (20%, including Physics, Chemistry, Biology, with an additional 100k completions in the PHD-Science subset), Math (15%, including Algebra, Calculus, Probability, with an additional 200k completions in kimiMath200k.jsonl), Computer Science (5%), Logical Questions (5%), Creative Writing (5%), and MultilingualSTEM (100k completions in MultilingualSTEM.jsonl). The total token count is 5B, and the dataset is suitable for text-generation and question-answering tasks, particularly for reasoning, chain-of-thought, instruction-tuning, and SFT applications. The data was collected using a modified Datagen by TeichAI over the course of about 80 hours.