jackf857/qwen3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-q_t-0.45-eta-0.1-s_star-0.6-margin-log
收藏资源简介:
该数据集是从一个名为New-DPO的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多个特征,如epoch(训练轮数)、step(步骤)、batch_size(批次大小)、mean(平均值)、std(标准差)、min(最小值)、p10(第10百分位数)、median(中位数)、p90(第90百分位数)、max(最大值)、pos_frac(正分数)、sample(每个示例的边际值)以及npy(保存的完整数组路径)。数据集来源于特定的模型训练运行,使用了Anthropic/hh-rlhf数据集作为混合器,并涉及特定的训练参数如beta、f_divergence_type等。
This dataset contains per-step margin summary statistics exported from a New-DPO training run. It includes various features from the training process such as epoch, step, batch_size, mean, std, min, p10, median, p90, max, pos_frac, sample (per-example margins), and npy (optional path to the saved full margin array). The dataset originates from a specific model training run, using the Anthropic/hh-rlhf dataset as a mixer, and involves specific training parameters like beta, f_divergence_type, etc.




