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

该数据集记录了New-DPO训练过程中每一步的边缘统计信息，包括epoch、step、batch_size、均值、标准差、最小值、百分位数、中位数、最大值、正样本比例等统计指标，以及每步有效批次的样本边缘值和保存的完整数组路径。数据来源于特定的模型训练运行，训练参数包括beta、f_divergence_type、s_star、eta等，数据集混合使用了Anthropic/hh-rlhf数据集。

This dataset contains per-step margin summary statistics exported from a New-DPO training run. It includes features such as epoch, step, batch_size, mean, std, min, percentiles, median, max, positive fraction, per-example margins for the effective batch, and optional paths to saved full margin arrays. The data comes from a specific model training run with parameters like beta, f_divergence_type, s_star, eta, and uses a dataset mixer with Anthropic/hh-rlhf.