知识卡片
标签选项帮助限定输出形式
分类指令明确列出允许标签,可减少同义措辞分散并方便解析。它不保证模型一定遵循,也不能替代标签定义和边界样本;训练与部署应使用一致的标签含义与映射。
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
Options clause. This introduces the LLM to the output space and can thus concentrate the probability mass over the defined label space. Without this guidance, the LLM would distribute its probability across several different tokens that express the same concept, for example there are several different ways of expressing the True label in a binary classification task.
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:摘录中关于标签选项帮助限定输出形式的定义、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:Options clause. This introduces the LLM to the output space and can thus concentrate the probability mass over the defined label space. Without this guidance, the LLM would distribute its probability across several different tokens that express the same concept, for example there are several different ways of expressing the True label in a binary classification task.