W-61/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.4-eta-0.5-margin-log

该数据集是从New-DPO训练运行中导出的每步边缘摘要统计数据集。数据集包含来自模型repo W-61/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.4-eta-0.5的训练运行结果，基础模型为W-61/llama-3-8b-base-sft-hh-helpful-4xh200。训练运行名称为llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.4-eta-0.5，并在W&B项目llama3-hh-new-dpo-hyperparamter-sweep中进行。数据集包含681行训练数据，每行数据包括epoch、step、batch_size、mean、std、min、p10、median、p90、max、pos_frac、sample和npy等列。训练参数包括beta=0.1、f_divergence_type=reverse_kl、f_alpha_divergence_coef=1.0、s_star=0.4、eta=0.5和q_t=0.45。数据集混合器使用了Anthropic/hh-rlhf数据集。

Per-step margin summary statistics exported from a New-DPO training run. The dataset contains training run results from the model repo W-61/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.4-eta-0.5, with the base model W-61/llama-3-8b-base-sft-hh-helpful-4xh200. The training run name is llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.45-s_star-0.4-eta-0.5, conducted in the W&B project llama3-hh-new-dpo-hyperparamter-sweep. The dataset includes 681 rows of training data, each containing columns such as epoch, step, batch_size, mean, std, min, p10, median, p90, max, pos_frac, sample, and npy. Training parameters include beta=0.1, f_divergence_type=reverse_kl, f_alpha_divergence_coef=1.0, s_star=0.4, eta=0.5, and q_t=0.45. The dataset mixer uses the Anthropic/hh-rlhf dataset.