W-61/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.4-eta-1-margin-log
收藏Hugging Face2026-04-28 更新2026-05-03 收录
下载链接：
https://hf-mirror.com/datasets/W-61/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.4-eta-1-margin-log
下载链接
链接失效反馈官方服务：
资源简介：
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。它包含了训练过程中的多个统计指标，如均值、标准差、最小值、百分位数等，以及每个步骤的样本边缘和可选的完整边缘数组保存路径。数据集基于Anthropic/hh-rlhf混合器，训练参数包括beta、f_divergence_type、s_star、eta等。
Per-step margin summary statistics exported from a New-DPO training run. It includes various statistical metrics during training such as mean, std, min, percentiles, etc., as well as per-example margins for the effective batch on each logged step and optional paths to saved full margin arrays. The dataset is based on the Anthropic/hh-rlhf mixer, with training arguments including beta, f_divergence_type, s_star, eta, etc.
提供机构：
W-61
