Ray2333/RiC_harmless_helpful
收藏Hugging Face2024-07-12 更新2024-07-13 收录
下载链接：
https://hf-mirror.com/datasets/Ray2333/RiC_harmless_helpful
下载链接
链接失效反馈官方服务：
资源简介：
hhrlhf数据集用于RiC训练，包含无害和有益的奖励机制。数据集中包含多个特征，如文本、提示、响应、输入ID等，并且这些特征的数据类型和结构也被详细列出。数据集的下载大小为653190564字节，数据集大小为1465037566字节，训练集包含291397个样本。输入ID是通过Llama2分词器获得的，奖励已经根据其对应的均值和标准差进行了归一化处理。
The hhrlhf dataset for RiC training with harmless and helpful rewards. It includes features like text, prompt, response, input_ids, query, two scores (score1 and score2), prompt with score, and prompt with score IDs. The dataset is divided into a training set with a large number of examples and bytes. The input_ids are generated from the Llama2 tokenizer, and the rewards are normalized according to their mean and standard deviation.
提供机构：
Ray2333
