slm-synthetic-educational-qa-mcq-general
收藏资源简介:
SLM Synthetic Educational Qa Mcq General 是一个合成的、非数学领域的、自包含的教育类多项选择题数据集,每个问题都附带解释。数据集于2026年5月30日生成,包含680,640条记录,存储大小为354.3 MB。数据以JSONL格式存储,每条记录是一个JSON对象,包含以下字段:`type`(类型)、`question`(问题)、`choices`(选项)、`correct_index`(正确答案索引)和`explanation`(解释)。数据集已进行精确去重。其预期用途包括用于小语言模型(SLM)的数据实验、预训练与继续预训练的混合训练、预训练以及行为评估。需要注意的是,数据是合成的,在使用前应进行检查,可能包含简单、重复或不完美的示例,不应被视为权威的事实知识来源。数据集基于MIT许可证发布。
SLM Synthetic Educational Qa Mcq General is a synthetic, non-mathematical, self-contained educational multiple-choice question dataset, with each question accompanied by an explanation. The dataset was generated on May 30, 2026, containing 680,640 records with a storage size of 354.3 MB. The data is stored in JSONL format, where each record is a JSON object with the following fields: `type` (type), `question` (question), `choices` (choices), `correct_index` (correct answer index), and `explanation` (explanation). The dataset has undergone exact deduplication. Its intended uses include data experiments for small language models (SLM), mixed training for pre-training and continued pre-training, pre-training, and behavioral evaluation. It should be noted that the data is synthetic and should be checked before use, as it may contain simple, repetitive, or imperfect examples and should not be considered an authoritative source of factual knowledge. The dataset is released under the MIT license.
数据集概述:SLM Synthetic Educational Qa Mcq General
基本信息
- 数据集名称:SLM Synthetic Educational Qa Mcq General
- 许可证:MIT License
- 语言:英语(en)
- 生成日期:2026-05-30
- 总记录数:680,640 条
- 存储大小:354.3 MB
标签
- synthetic(合成数据)
- llm(大语言模型)
- pretraining(预训练)
- reasoning(推理)
- educational(教育)
文件结构
数据集仅包含一个JSONL格式文件:
| 文件名 | 记录数 | 大小 |
|---|---|---|
educational_qa_mcq_general.jsonl |
680,640 | 354.3 MB |
记录格式
每条记录为一行JSON对象,字段结构如下:
- type:数据类型
- question:问题内容
- choices:选项列表
- correct_index:正确答案索引
- explanation:答案解析
去重处理
- 已执行精确去重(exact-deduplication)
- 未使用模糊MinHash去重,因为合成示例通常设计上共享有用结构
预期用途
- SLM(小语言模型)数据实验
- 预训练或继续预训练混合数据
- 行为评估
局限性
- 数据为合成生成,使用前应进行检查
- 可能包含简单、重复或不完美的示例
- 不应被视作权威事实知识的来源




