W-61/llama-3-8b-base-margin-dpo-hh-helpful-margin-log
收藏资源简介:
# W-61/llama-3-8b-base-margin-dpo-hh-helpful-margin-log Per-step margin summary statistics exported from a margin-DPO training run. ## Source Run - Model repo id: `W-61/llama-3-8b-base-margin-dpo-hh-helpful-8xh200` - Base model: `W-61/llama-3-8b-base-sft-hh-helpful-8xh200` - Run name: `llama-3-8b-base-margin-dpo-hh-helpful-8xh200-20260410-172009` - Margin log path: `/scratch/feng.yulu/dynamic-dpo-v4/outputs/llama-3-8b-base-margin-dpo-hh-helpful-8xh200-20260410-172009/margin_logs` - Published split: `train` - Rows: `340` ## Columns - `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) - `npy` (optional path to the saved full margin array when `margin_save_full=true`) ## Dataset Mixer ```json { "Anthropic/hh-rlhf": 1.0 } ```
# W-61/llama-3-8b-base-margin-dpo-hh-helpful-margin-log 本数据集为从某次边际DPO(Margin-DPO)训练流程中导出的逐步边际统计摘要。 ## 训练运行源信息 - 模型仓库标识符:`W-61/llama-3-8b-base-margin-dpo-hh-helpful-8xh200` - 基础模型:`W-61/llama-3-8b-base-sft-hh-helpful-8xh200` - 训练运行名称:`llama-3-8b-base-margin-dpo-hh-helpful-8xh200-20260410-172009` - 边际日志路径:`/scratch/feng.yulu/dynamic-dpo-v4/outputs/llama-3-8b-base-margin-dpo-hh-helpful-8xh200-20260410-172009/margin_logs` - 发布数据集划分:`train`(训练集) - 数据总行数:`340` ## 字段说明 - `epoch`(训练轮次) - `step`(训练步数) - `batch_size`(批次大小) - `mean`(均值) - `std`(标准差) - `min`(最小值) - `p10`(10%分位数) - `median`(中位数) - `p90`(90%分位数) - `max`(最大值) - `pos_frac`(正样本占比) - `sample`:该日志步数下有效批次的逐示例边际值 - `npy`:当`margin_save_full=true`时,保存的完整边际数组的可选路径 ## 数据集混合配置 json { "Anthropic/hh-rlhf": 1.0 }




