PJMixers-Dev/foundRP-qwq-all-kcpp

foundRP-qwq-all-kcpp数据集是一个用于训练对话模型的数据集。在训练过程中，应遮蔽除最后一轮对话之外的所有内容，因为该数据集的设计目的是教会模型无论输入内容如何都进行思考并输出。数据集遵循R1模型的训练方式。

The foundRP-qwq-all-kcpp dataset is a conversational dataset for training dialogue models. During training, everything except the last turn of the conversation should be masked, as the dataset is designed to teach the model to always output thinking regardless of the input. The dataset is set up to be trained in the style of the R1 model.