Dumoura/oulipo_dpo_nano
收藏官方服务：
资源简介：
OULIPO DPO数据集是一个直接偏好优化（DPO）的数据集，用于训练语言模型生成遵循OULIPO文学约束（如 lipograms、回文、univocalisms、数学模式等）的创意写作。数据集包含205个偏好对，采用DPO格式，并分为164对的训练集和41对的测试集。数据集的评分范围、数据划分细节、数据结构以及OULIPO约束的多样性都有详细的描述。
The OULIPO DPO Dataset is a Direct Preference Optimization (DPO) dataset designed for training language models to generate creative writing based on OULIPO literary constraints (such as lipograms, palindromes, univocalisms, mathematical patterns, etc.). The dataset consists of 205 preference pairs in the DPO format, split into a training set of 164 pairs and a test set of 41 pairs. The score ranges, data split details, dataset structure, and diversity of OULIPO constraints are all thoroughly described.
提供机构：
Dumoura
