The LMSYS-USP dataset contains high-quality dialogues with inferred user profiles(provide natural descriptions encompassing both objective facts and subjective characteristics), generated through a two-stage profiling pipeline (see our paper for details). The dataset includes a training set (87,882 examples), a validation set (4,626)… See the full description on the dataset page: https://huggingface.co/datasets/wangkevin02/LMSYS-USP.
