amd/ReasonLite-Dataset
收藏Hugging Face2026-01-22 更新2025-12-20 收录
下载链接：
https://hf-mirror.com/datasets/amd/ReasonLite-Dataset
下载链接
链接失效反馈官方服务：
资源简介：
ReasonLite数据集包含从Polaris和OpenMathReasoning收集的343K数学问题。使用GPT-OSS作为教师模型，在中高推理模式下生成了9.1M原始答案，并通过多数投票生成伪标签，最终保留了6.1M样本。数据集分为Short CoT（4.3M）和Long CoT（1.8M）两部分。
The ReasonLite dataset consists of 343K math problems collected from Polaris and OpenMathReasoning. Using GPT-OSS as the teacher model, 9.1M raw answers were generated under medium and high reasoning modes. Pseudo-labels were produced via majority voting, and 6.1M samples were retained. The dataset is divided into Short CoT (4.3M) and Long CoT (1.8M) parts.
提供机构：
amd
