RussianNLP/repa
收藏官方服务：
资源简介：
REPA是一个俄语语言数据集，包含1003个经过分类的用户查询和六个开源指令微调的LLMs的回答。数据集提供了针对10种错误类型和总体印象的细粒度人工偏好标注。
REPA is a Russian language dataset consisting of 1,003 categorized user queries and responses from six open-source instruction-finetuned Russian LLMs, with fine-grained human preference annotations for ten error types and overall impression.
提供机构：
RussianNLP
