MWilinski/hh-rlhf-helpful-online

HH-RLHF Helpful Online数据集是一个处理过的变体，基于Anthropic的HH-RLHF数据目录中的helpful-online部分。它将原始字符串转换为适合TRL偏好和RM训练管道的对话聊天格式。处理过程保留了人与助手的对话回合，并且明确区分了选中的和被拒绝的对话延续。数据集分为训练集和测试集。

This dataset is a processed variant of the Anthropic HH-RLHF data directory `helpful-online`. It converts the raw strings into conversational chat format that is compatible with the TRL preference and RM training pipelines. The transformation preserves the human and assistant turns while clearly separating the chosen and rejected continuations. The dataset is split into `train` and `test` sets.