Skywork/Skywork-Reward-Preference-80K-v0.1
收藏Hugging Face2024-10-25 更新2025-04-08 收录
下载链接:
https://hf-mirror.com/datasets/Skywork/Skywork-Reward-Preference-80K-v0.1
下载链接
链接失效反馈官方服务:
资源简介:
Skywork Reward Preference 80K是一个包含80K个偏好对的数据集,用于训练Skywork-Reward-Gemma-2-27B和Skywork-Reward-Llama-3.1-8B模型。这个数据集从多个公开数据源中精心挑选了高质量的偏好对,并针对特定的能力和知识领域进行了构建。在构建数据集时,采用了一些技巧来提高性能并保持各领域的平衡。
Skywork Reward Preference 80K is a dataset of 80K preference pairs used for training the Skywork-Reward-Gemma-2-27B and Skywork-Reward-Llama-3.1-8B models. It is carefully curated from multiple publicly available data sources to include high-quality preference pairs and is tailored for specific capability and knowledge domains. Some techniques were adopted during dataset construction to improve performance and maintain balance across domains.
提供机构:
Skywork



