知识卡片
指令演化分开难度与广度
增加约束、步骤或输入复杂度可沿深度演化;增加同领域任务或主题可沿广度演化。演化后要检查任务是否仍可解、答案是否匹配和范围是否漂移,再纳入训练,不把复杂措辞直接等同有益难题。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 微调
原书出处 · 1 处
- · 查看原书依据
Instruction evolution: The seed examples are evolved using in-depth and in-breadth strategies. In-depth evolution increases the complexity and difficulty of the original instruction through five types of prompts: Adding constraints Increasing reasoning steps Asking deeper questions Asking more specific questions Increasing the complexity of the input In-breadth evolution increases topic coverage by generating a completely new instruction from the same domain as the original instruction.
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:摘录中关于指令演化分开难度与广度的定义、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:Instruction evolution: The seed examples are evolved using in-depth and in-breadth strategies. In-depth evolution increases the complexity and difficulty of the original instruction through five types of prompts: Adding constraints Increasing reasoning steps Asking deeper questions Asking more specific questions Increasing the complexity of the input In-breadth evolution increases topic coverage by generating a completely new instruction from the same domain as the original instruction.