yufan/SFT_OpenO1_Scored
收藏Hugging Face2024-12-17 更新2024-12-21 收录
下载链接：
https://hf-mirror.com/datasets/yufan/SFT_OpenO1_Scored
下载链接
链接失效反馈官方服务：
资源简介：
该数据集由以下开源SFT数据集组成：[`OpenO1-SFT`](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT)。数据清理方法包括：1. 仅保留使用FastText语言检测的特定语言（EN/DE/ES/ZH/IT/JA/FR）；2. 去除重复数据，确保每个提示只出现一次；3. 移除响应长度超过8K字符的数据；4. 使用奖励模型（Mistral-Nemo-Base-2407在偏好数据集上微调）对每个数据点进行评分。
This dataset consists of multiple open-sourced SFT datasets, featuring elements such as prompt, thought, response, and a nested messages feature. The dataset has undergone various cleaning methods, including language detection, deduplication, response length limitation, and scoring using a reward model. The dataset is divided into a training set containing 125494 samples. The source of the dataset is OpenO1-SFT.
提供机构：
yufan
