JAMhunggingface/Natural-Reasoning-gpt-oss-120B-S1

这是一个专为高效知识蒸馏任务设计的指令微调数据集，基于大规模推理语料库facebook/natural_reasoning的前10万个问题构建。数据集旨在通过监督式微调（SFT）范式，将教师模型gpt-oss-120-high的高级、多步骤推理能力高保真地迁移至学生模型。数据集包含问题、答案和由教师模型生成的思维链（CoT_reasoning），以及基础数据集提供的参考答案。数据集以其高质量、高难度和多样性著称，覆盖STEM、经济学、社会科学等多个领域。

This is a meticulously curated instruction fine-tuning dataset designed specifically for efficient knowledge distillation tasks. Built upon the first 100,000 questions from the large-scale reasoning corpus facebook/natural_reasoning, it aims to transfer the advanced, multi-step reasoning capabilities of the teacher model gpt-oss-120-high to a student model with high fidelity through a Supervised Fine-Tuning (SFT) paradigm. The dataset includes questions, answers, and the complete Chain-of-Thought (CoT) reasoning process generated by the teacher model, along with reference answers from the base dataset. It is renowned for its high quality, difficulty, and diversity, covering multiple domains such as STEM, economics, and social sciences.