long-retrieval
收藏资源简介:
该数据集是一个用于语言模型训练的多任务对话数据集,包含两个独立的配置子集:instruct-30s和reasoning-30s。instruct-30s子集包含743个训练样本,每个样本由一组对话交互组成,包括用户查询(query)和对应的回答(answer)。reasoning-30s子集包含680个训练样本,每个交互不仅包含查询(query)和最终回答(answer),还包含一个中间推理步骤(think),模拟了逐步思考的过程。数据集以文本字符串形式存储,总数据量约为40MB,适用于训练或微调语言模型进行指令理解、对话生成以及思维链推理等任务。
This dataset is a multi-task dialogue dataset for language model training, consisting of two independent configuration subsets: instruct-30s and reasoning-30s. The instruct-30s subset contains 743 training samples, each composed of a set of dialogue interactions including a user query and the corresponding answer. The reasoning-30s subset contains 680 training samples, where each interaction includes not only the query and final answer but also an intermediate reasoning step (think), simulating a step-by-step thinking process. The dataset is stored in text string format, with a total data volume of approximately 40MB, and is suitable for training or fine-tuning language models for tasks such as instruction understanding, dialogue generation, and chain-of-thought reasoning.
数据集概述:ReactiveAI/long-retrieval
- 数据集地址:https://huggingface.co/datasets/ReactiveAI/long-retrieval
配置与结构
该数据集包含两个配置(config):
| 配置名称 | 特征字段 | 数据集大小(字节) | 样本数量 | 训练集本地路径 |
|---|---|---|---|---|
instruct-30s |
interactions 列表,包含 answer 和 query(均为字符串) |
4,392,681 | 743 | instruct-30s/train-* |
reasoning-30s |
interactions 列表,包含 query、think、answer(均为字符串) |
35,988,272 | 680 | reasoning-30s/train-* |
数据集拆分
- 两个配置均仅包含
train拆分,无验证集或测试集。 - 每个配置的下载大小与数据集大小基本一致(
instruct-30s下载大小约 2.65 MB,reasoning-30s下载大小约 36.18 MB)。
数据特征
instruct-30s:每条数据包含一段交互记录,由query(查询)和answer(回答)组成。reasoning-30s:每条数据包含一段交互记录,由query(查询)、think(思考过程)和answer(回答)组成,提供了更丰富的推理相关信息。




