W-61/llama-multi-beta-testing-margin_dpo-hh-helpful-beta-0p3-20260429-085449-margin
收藏Hugging Face2026-04-29 更新2026-05-03 收录
下载链接：
https://hf-mirror.com/datasets/W-61/llama-multi-beta-testing-margin_dpo-hh-helpful-beta-0p3-20260429-085449-margin
下载链接
链接失效反馈官方服务：
资源简介：
这个数据集是从一个Margin-DPO训练运行中导出的每步边际摘要统计。它包含了训练过程中的各种统计指标，如epoch、step、batch_size、mean、std、min、p10、median、p90、max、pos_frac等，以及每个步骤的边际样本数据（sample列）。数据集来源于Anthropic/hh-rlhf，训练参数包括beta=0.3，f_divergence_type=reverse_kl，f_alpha_divergence_coef=1.0。数据集共有681行，仅包含训练集。
Per-step margin summary statistics exported from a Margin-DPO training run. It includes various statistical metrics during training such as epoch, step, batch_size, mean, std, min, p10, median, p90, max, pos_frac, and per-example margins for the effective batch on each logged step (sample column). The dataset is sourced from Anthropic/hh-rlhf, with training parameters including beta=0.3, f_divergence_type=reverse_kl, f_alpha_divergence_coef=1.0. The dataset contains 681 rows and only includes the training split.
提供机构：
W-61
