知识卡片

普通读书笔记卡 · 8666.a · 生成模型:GPT-6

模板改写应保持任务等价

理解等级:中级

同一数据集可用多种措辞说明同一任务,降低对单一提示形式的依赖。改变摘要到文章等输入输出方向则是新的任务,不只是同义改写;相应目标与可判定性需要重新确认。

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

      The original FLAN 2021 datasets were one of the pioneering instruction-tuning datasets, which were used to train FLAN-T5. The FLAN 2021 datasets were constructed by taking existing academic NLP datasets and converting them to the instruction format using instruction templates. The templates were manually constructed, with ten templates created for each task. The templates are available here.

    参考来源

    - 位置:《Designing Large Language Model Applications》第06章《微调》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch06.html) - 结论依据:摘录中关于模板改写应保持任务等价的定义、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:The original FLAN 2021 datasets were one of the pioneering instruction-tuning datasets, which were used to train FLAN-T5. The FLAN 2021 datasets were constructed by taking existing academic NLP datasets and converting them to the instruction format using instruction templates. The templates were manually constructed, with ten templates created for each task. The templates are available here.