遇见数据集

Ethosoft/OpenR1TurkishReasoning

收藏
Hugging Face2026-05-26 更新2026-05-31 收录
官方服务:

资源简介:

OpenR1-Math-220k Türkçe是土耳其语中第一个且最全面的数学推理数据集,包含93,733个训练样本。该数据集是open-r1/OpenR1-Math-220k数据集中default分割的训练数据的土耳其语翻译版本,填补了土耳其语数学推理领域的关键空白。数据集涵盖代数、几何、数论、微积分、组合数学、预微积分和中级代数等多个数学领域,采用链式思维推理方法。每个样本包含以下字段:问题(土耳其语)、解决方案/推理轨迹(土耳其语)、最终答案、问题类型、问题格式、原始来源、消息(适用于监督微调的聊天格式)、由DeepSeek R1生成的推理轨迹以及数学验证正确性标签。数据集旨在促进土耳其语数学大型语言模型、土耳其语STEM教育工具和多语言推理模型的开发,覆盖从奥林匹克竞赛题目、竞赛数学到教科书问题的广泛难度范围。数据集基于Apache 2.0许可证发布。

OpenR1-Math-220k Türkçe is the first and most comprehensive mathematical reasoning dataset in Turkish, containing 93,733 training examples. It is a Turkish translation of the training data from the default split of the open-r1/OpenR1-Math-220k dataset, addressing a critical gap in Turkish mathematical reasoning. The dataset covers various mathematical domains including algebra, geometry, number theory, calculus, combinatorics, precalculus, and intermediate algebra, utilizing chain-of-thought reasoning. Each example includes the following fields: problem (in Turkish), solution/reasoning trace (in Turkish), final answer, problem type, question format, original source, messages (chat-formatted for supervised fine-tuning), reasoning traces generated by DeepSeek R1, and mathematical correctness verification labels. The dataset is designed to advance the development of Turkish mathematical LLMs, Turkish STEM educational tools, and multilingual reasoning models, encompassing a wide range of difficulties from olympiad problems, competition mathematics, to textbook exercises. It is released under the Apache 2.0 license.

提供机构:
Ethosoft
二维码
社区交流群
二维码
科研交流群
商业服务