jackf857/llama-3-8b-base-scheduled-beta-margin-dpo-hh-harmless-margin-log
收藏数据链接:
官方服务:
资源简介:
该数据集是从一个margin-DPO训练运行中导出的每步边际摘要统计,包含661行训练数据。数据列包括epoch、step、batch_size、mean、std、min、p10、median、p90、max、pos_frac、sample(每个示例在记录步骤的有效批次中的边际)以及npy(当margin_save_full=true时保存的完整边际数组的可选路径)。数据集混合器配置显示使用了Anthropic/hh-rlhf数据集,权重为1.0。
Per-step margin summary statistics exported from a margin-DPO training run, containing 661 rows of training data. Columns include epoch, step, batch_size, mean, std, min, p10, median, p90, max, pos_frac, sample (per-example margins for the effective batch on that logged step), and npy (optional path to the saved full margin array when margin_save_full=true). The dataset mixer configuration shows the use of the Anthropic/hh-rlhf dataset with a weight of 1.0.
提供机构:
jackf857


