jackf857/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.4-margin

该数据集是从一个名为jackf857/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.4的新DPO训练运行中导出的每一步边际摘要统计。数据集包含训练过程中的各种统计指标，如epoch、step、batch_size、mean、std、min、p10、median、p90、max、pos_frac、sample（每个示例的边际）和npy（保存完整边际数组的可选路径）。训练运行基于模型repo id: jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128，使用了HuggingFaceH4/ultrafeedback_binarized数据集进行混合训练。训练参数包括beta: 0.01、f_divergence_type: reverse_kl、f_alpha_divergence_coef: 1.0、s_star: 0.4、eta: 0.1和q_t (q_target): 0.45。

Per-step margin summary statistics exported from a New-DPO training run named jackf857/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.4. The dataset includes various statistical metrics during training, 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 training run is based on the model repo id: jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128 and uses the HuggingFaceH4/ultrafeedback_binarized dataset for mixed training. Training parameters include beta: 0.01, f_divergence_type: reverse_kl, f_alpha_divergence_coef: 1.0, s_star: 0.4, eta: 0.1, and q_t (q_target): 0.45.