jz666/gemma2-ultrafeedback-ppl

该数据集包含了对话生成的相关数据，具体字段包括：prompt_id（提示ID），prompt（提示文本），all_generated_responses（所有生成的响应），all_rm_scores（所有响应的RM分数），chosen（选定的响应及其角色），rejected（被拒绝的响应及其角色），ppl_chosen（选定响应的PPL值），nll_chosen（选定响应的NLL值），ppl_rejected（被拒绝响应的PPL值），nll_rejected（被拒绝响应的NLL值）。数据集分为训练集和测试集，训练集包含59569个示例，大小为792097911字节，测试集包含1941个示例，大小为26136876字节。

The dataset includes dialogue generation related data, with specific fields such as prompt_id (prompt ID), prompt (prompt text), all_generated_responses (all generated responses), all_rm_scores (RM scores for all responses), chosen (selected response and its role), rejected (rejected response and its role), ppl_chosen (PPL value of the selected response), nll_chosen (NLL value of the selected response), ppl_rejected (PPL value of the rejected response), and nll_rejected (NLL value of the rejected response). The dataset is divided into training and test sets, with the training set containing 59569 examples and a size of 792097911 bytes, and the test set containing 1941 examples and a size of 26136876 bytes.