JayHyeon/trl_ultrainteract-pair

UltraInteract Pair数据集是一个经过处理的用户助手交互数据集，采用标准格式，适用于偏好学习。该数据集源自openbmb/UltraInteract_pair数据集，旨在用于微调及评估模型在复杂推理的对齐任务中的表现。数据集包含三个列：提示文本（prompt）、首选响应（chosen）和非首选响应（rejected），遵循TRL偏好格式。

The UltraInteract Pair dataset is a processed dataset of user-assistant interactions in standard format for preference learning. Derived from the openbmb/UltraInteract_pair dataset, it is designed for fine-tuning and evaluating models in alignment tasks, especially for complex reasoning. The dataset includes three columns: text input prompt (prompt), preferred response (chosen), and non-preferred response (rejected), following the TRL preference format.