WildChat-long-SMAT
收藏资源简介:
该数据集包含两个配置:instruct-30s和reasoning-30s,均设计用于语言模型训练。instruct-30s配置包含1728个训练样本,每个样本由一对字符串字段组成:query(查询)和answer(回答),适用于指令跟随任务的微调或评估。reasoning-30s配置包含728个训练样本,每个样本包含三个字符串字段:query(查询)、think(思考过程)和answer(回答),适用于思维链推理任务的模型训练或评估。数据以文本形式存储,支持模型在理解和生成响应方面的能力提升。
This dataset includes two configurations: instruct-30s and reasoning-30s, both developed for language model training. The instruct-30s configuration contains 1728 training samples, each composed of a pair of string fields: query and answer, and is applicable to the fine-tuning or evaluation of instruction-following tasks. The reasoning-30s configuration consists of 728 training samples, each with three string fields: query, think, and answer, and is tailored for model training or evaluation of chain-of-thought reasoning tasks. The data is stored in text format, supporting the enhancement of models' capabilities in understanding and generating responses.
数据集名称
WildChat-long-SMAT
数据集配置
该数据集包含两个配置,每个配置均仅包含训练集:
1. instruct-30s
- 样本数量:1,728条
- 数据大小:约137.33 MB(下载大小约56.23 MB)
- 特征字段:
interactions(列表,包含两项):answer(字符串):回答内容query(字符串):用户提问
2. reasoning-30s
- 样本数量:988条
- 数据大小:约135.91 MB(下载大小约135.89 MB)
- 特征字段:
interactions(列表,包含三项):query(字符串):用户提问think(字符串):推理过程answer(字符串):最终回答
数据文件
所有数据文件均位于对应配置目录下,文件路径模式为:
instruct-30s/train-*reasoning-30s/train-*




