yunjae-won/Qwen3-30B-MagpieLM-SFT-Outputs-v0.1-shard1-llama8b_idx0
收藏Hugging Face2025-10-18 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/yunjae-won/Qwen3-30B-MagpieLM-SFT-Outputs-v0.1-shard1-llama8b_idx0
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了一个提示（prompt）字段和三个与之相关的选择字段：选中的答案（chosen）、被选中的答案的教师对数概率（chosen_teacher_logp）、被拒绝的答案（rejected）以及两个与被拒绝答案相关的概率字段：学生拒绝的对数概率（rejected_student_logp）和教师拒绝的对数概率（rejected_teacher_logp）。数据集仅包含训练集，共有50000个样本，总大小为775253016字节。
The dataset includes a prompt field and three associated choice fields: the chosen answer (chosen), the log probability of the chosen answer by the teacher (chosen_teacher_logp), the rejected answer (rejected), and two probability fields associated with the rejected answer: the students log probability of rejection (rejected_student_logp) and the teachers log probability of rejection (rejected_teacher_logp). The dataset contains only a training set with 50,000 samples, totaling 775253016 bytes in size.
提供机构：
yunjae-won
