jz666/gemma2-ultrafeedback-split-length

该数据集包含以下字段：prompt_id、prompt、all_generated_responses、all_rm_scores、chosen（包括content和role）、rejected（包括content和role）、ppl_chosen、nll_chosen、ppl_rejected和nll_rejected。数据集分为训练集、测试集以及不同长度的训练集子集，具体包括：train、test、train_len_q1_shortest、train_len_q2、train_len_q3、train_len_q4和train_len_q5_longest。训练集大小为792097911字节，包含59569个样本；测试集大小为26136876字节，包含1941个样本。不同长度的训练集子集包含样本数量略有不同。数据集总下载大小为815056280字节，总数据大小为1610332698字节。

The dataset includes the following fields: prompt_id, prompt, all_generated_responses, all_rm_scores, chosen (including content and role), rejected (including content and role), ppl_chosen, nll_chosen, ppl_rejected, and nll_rejected. The dataset is split into training set, test set, and subsets of training sets with different lengths, specifically: train, test, train_len_q1_shortest, train_len_q2, train_len_q3, train_len_q4, and train_len_q5_longest. The training set is 792097911 bytes in size and contains 59569 samples; the test set is 26136876 bytes in size and contains 1941 samples. The subsets of training sets with different lengths contain slightly different numbers of samples. The total download size of the dataset is 815056280 bytes, and the total dataset size is 1610332698 bytes.