W-61/llama-multi-beta-testing-margin_dpo-hh-helpful-beta-0p8-20260429-085449-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-0p8-20260429-085449-margin
下载链接
链接失效反馈官方服务：
资源简介：
这是一个从Margin-DPO训练运行中导出的每步边际摘要统计数据。数据集包含了训练过程中的各种统计指标，如均值、标准差、最小值、百分位数等，以及每个步骤的样本边际。数据来源于一个特定的训练运行，使用了Anthropic/hh-rlhf数据集作为混合器，训练参数包括beta值为0.8，f_divergence_type为reverse_kl等。
Per-step margin summary statistics exported from a Margin-DPO training run. The dataset includes various statistical metrics from the training process, such as mean, standard deviation, minimum, percentiles, etc., as well as per-step sample margins. The data comes from a specific training run using the Anthropic/hh-rlhf dataset as a mixer, with training parameters including a beta value of 0.8, f_divergence_type of reverse_kl, etc.
提供机构：
W-61
