W-61/llama-3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-s_star-0.4-eta-0.1-q_t-0.5-margin-log

该数据集是从一个New-DPO训练运行中导出的每步边缘摘要统计信息。数据集包含多个特征，如epoch（训练轮数）、step（步数）、batch_size（批次大小）、mean（平均值）、std（标准差）、min（最小值）、p10（第10百分位数）、median（中位数）、p90（第90百分位数）、max（最大值）、pos_frac（正分数）、sample（每个示例的边缘）和npy（保存完整边缘数组的可选路径）。数据集来源于模型库ID为W-61/llama-3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-s_star-0.4-eta-0.1-q_t-0.5的训练运行，使用了Anthropic/hh-rlhf数据集进行混合。This dataset comprises per-step marginal summary statistics exported from a New-DPO training run. It includes multiple features including epoch (training epoch), step (training step), batch_size (batch size), mean (average value), std (standard deviation), min (minimum value), p10 (10th percentile), median (median value), p90 (90th percentile), max (maximum value), pos_frac (positive fraction), sample (marginal of each example), and npy (optional path for saving full marginal arrays). This dataset originates from the training run with model repository ID W-61/llama-3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-s_star-0.4-eta-0.1-q_t-0.5, which was mixed with the Anthropic/hh-rlhf dataset.