Aratako/Synthetic-JP-Preference-Dataset-Qwen2.5_72B-191k
收藏Hugging Face2025-02-02 更新2025-02-15 收录
下载链接：
https://hf-mirror.com/datasets/Aratako/Synthetic-JP-Preference-Dataset-Qwen2.5_72B-191k
下载链接
链接失效反馈官方服务：
资源简介：
这是一个包含190854个样本的日本语合成Preference数据集，使用5种开放模型和Qwen2.5-72B-Instruct-GPTQ-Int8模型生成。数据集包含id，messages（包括content和role），instruction，chosen，rejected等字段，并使用Qwen2.5-72B-Instruct-GPTQ-Int8模型进行judge，考虑了位置偏差的影响。数据集的license主要是Apache 2.0，但在特定情况下可能需要遵守Qwen License。
This is a Japanese synthetic Preference dataset containing 190854 samples, generated using five open models and the Qwen2.5-72B-Instruct-GPTQ-Int8 model. The dataset includes fields such as id, messages (including content and role), instruction, chosen, rejected, and uses the Qwen2.5-72B-Instruct-GPTQ-Int8 model for judge, considering the influence of positional bias. The datasets license is primarily Apache 2.0, but may be subject to the Qwen License under certain circumstances.
提供机构：
Aratako
