jz666/gemma2-ultrafeedback-ppl-split

该数据集包含多个特征，如提示ID、提示文本、所有生成的响应、所有RM评分、选择的响应及其角色、拒绝的响应及其角色、选择的响应的 perplexity 和负对数似然等。数据集分为训练集、测试集、过滤后的完整训练集以及根据特定标准划分的其他训练集。没有明确的数据集用途或内容的描述。

The dataset includes multiple features such as prompt ID, prompt text, all generated responses, all RM scores, chosen response and its role, rejected response and its role, perplexity and negative log-likelihood of the chosen response, etc. The dataset is split into training set, test set, filtered full training set, and other training sets based on specific criteria. There is no explicit description of the datasets purpose or content.