Rich740804/OP-V2
收藏资源简介：
OP-V2数据集是一个基于ValuePrism数据集改进和扩充的版本，旨在用于训练和评估能够对复杂社会情境生成多样化、多角度回应的模型。数据集包括三个子集：主训练集(train)，包含具有多个不同视角的情境-回应对；测试集(test)，根据人类参考视角的数量分为七个子任务；以及总结精调集(summary_sft)，格式化为对总结性多角度输出的监督微调。
The OP-V2 dataset is a refined and augmented version based on the ValuePrism dataset, designed for training and evaluating models capable of generating diverse, multi-perspective responses to complex social situations. The dataset includes three subsets: the main training set (train) with situation-response pairs of multiple distinct perspectives; the evaluation set (test) partitioned into seven sub-tasks based on the number of human reference perspectives; and the summary fine-tuning set (summary_sft) formatted for supervised fine-tuning on summarized multi-perspective outputs.
