got-oss-120b 蒸馏 STEM 工科推理数据集

GPT-OSS-120B-Distilled-Reasoning-STEM 是一个基于 gpt-oss-120b 模型蒸馏生成的推理型数据集，涵盖数学、物理、化学、计算机科学、工程与生命科学等多个 STEM 领域。数据以 JSON Lines 格式存储，包含 input（题目）、reasoning（推理链）和 answer（最终答案），能够支持链式思维训练（CoT）与常规问答任务，适合用于研究和提升模型的多步推理与跨学科问题求解能力。

GPT-OSS-120B-Distilled-Reasoning-STEM is a reasoning-oriented dataset distilled from the gpt-oss-120b model. It covers multiple STEM disciplines including mathematics, physics, chemistry, computer science, engineering and life sciences. The data is stored in JSON Lines format, containing three core fields: "input" (question), "reasoning" (chain of thought), and "answer" (final answer). This dataset can support both chain-of-thought (CoT) training and conventional question answering tasks, and is suitable for researching and enhancing models' multi-step reasoning and interdisciplinary problem-solving capabilities.

搜集汇总

背景与挑战

背景概述

该数据集基于gpt-oss-120b模型生成，专注于STEM领域的推理和问题解决，以JSONL格式提供，明确区分推理链和最终答案，支持COT/SFT训练和评估。它覆盖数学、物理等多个学科，但存在生成偏差和学科覆盖不均的局限性，采用CC-BY-4.0许可证，适用于学术和工业场景。