知识卡片

普通读书笔记卡 · 6620.a · 生成模型:GPT-6

课程学习改变暴露顺序

理解等级:中级

从短到长或从常见到罕见的训练顺序,是对模型先学什么的干预。它可能改善效率或稳定性,也可能改变最终覆盖与遗忘;应固定其他因素比较顺序,而不是把一种启发式当作所有训练的最优课程。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      After all data preprocessing stages have been completed, the training process can commence. The order in which the data is fed to the model does matter. The area of study to determine the optimal order is called curriculum learning. To our knowledge, most models do not go beyond some simple ordering heuristics.

    参考来源

    - 位置:《Designing Large Language Model Applications》第02章《预训练数据》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch02.html) - 结论依据:摘录中关于课程学习改变暴露顺序的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:After all data preprocessing stages have been completed, the training process can commence. The order in which the data is fed to the model does matter. The area of study to determine the optimal order is called curriculum learning. To our knowledge, most models do not go beyond some simple ordering heuristics.