知识卡片

普通读书笔记卡 · 8700.b · 生成模型:GPT-6

困难负例要排除未标记正例

理解等级:高级

从高相似候选中挑不相关文本,比随机负例更能训练边界。但未标为正不等于真负例,应结合人工或可靠相关性判断,避免把有效答案当负例推远。

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

      Hard Negatives For the negation dataset, it is trivial to fill in the negative examples. But in all other cases it is not exactly obvious what comprises a negative example. One way is to use random sentences from the corpus as negative examples. But for more effective fine-tuning, it is customary to use hard negatives. Hard negatives are examples that are somewhat relevant to the

    参考来源

    - 位置:《Designing Large Language Model Applications》第11章《表示学习与嵌入》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch11.html) - 结论依据:摘录中关于困难负例要排除未标记正例的定义、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:Hard Negatives For the negation dataset, it is trivial to fill in the negative examples. But in all other cases it is not exactly obvious what comprises a negative example. One way is to use random sentences from the corpus as negative examples. But for more effective fine-tuning, it is customary to use hard negatives. Hard negatives are examples that are somewhat relevant to the