got-oss-120B 蒸馏推理数据集

使用满血gpt-oss-120B推理模型，蒸馏生成的答案与链式思维（Chain-of-Thought，CoT），适用于训练 LLM 理解与生成复杂推理过的文本响应。经过清洗过滤与对齐，数据规模小于 10,000 条，易于载入和实验，适合教学、调试或作为其他数据集的补充。

This dataset consists of distilled answers and Chain-of-Thought (CoT) samples generated via the full-capacity GPT-OSS-120B inference model, and is designed for training Large Language Models (LLMs) to comprehend and generate text responses involving complex reasoning processes. Following cleaning, filtering and alignment processing, the dataset contains fewer than 10,000 samples. It is easy to load and conduct experiments on, making it suitable for educational purposes, debugging tasks, or serving as a supplementary dataset for other training resources.

搜集汇总

背景与挑战

背景概述

该数据集名为GPT-oss-120B-Distilled-Reasoning-math，基于gpt-oss-120b模型生成，专注于数学问题求解任务，采用JSON Lines格式存储。数据集提供了高质量的推理过程，包含详细的逻辑链和LaTeX数学表达式，并支持多种结构化模板以适应不同训练需求。