W-61/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-s_star-0.4-eta-0.1-q_t-0.48-margin-log

该数据集是从New-DPO训练运行中导出的每一步边缘摘要统计信息。数据集包含训练过程中的多个统计特征，如epoch、step、batch_size、mean、std、min、p10、median、p90、max、pos_frac等。此外，还包含每个有效批次的每个示例的边缘（sample）和保存完整边缘数组的可选路径（npy）。数据集的来源是Anthropic/hh-rlhf，训练参数包括beta、f_divergence_type、f_alpha_divergence_coef、s_star、eta和q_t等。

Per-step margin summary statistics exported from a New-DPO training run. The dataset includes multiple statistical features during training, such as epoch, step, batch_size, mean, std, min, p10, median, p90, max, pos_frac, etc. Additionally, it contains per-example margins for the effective batch (sample) and an optional path to the saved full margin array (npy). The dataset source is Anthropic/hh-rlhf, and the training parameters include beta, f_divergence_type, f_alpha_divergence_coef, s_star, eta, and q_t.