知识卡片

普通读书笔记卡 · 8700.a · 生成模型:GPT-6

三元组标签表达任务相关性

理解等级:中级

anchor、positive和negative指定当前目标下什么应更近。负例未必文字不相似,也未必全局错误;例如主题检索和矛盾检索可能对同一对句子采用不同标签。

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

      anchor, positive, negative) examples. For a given anchor sentence, the positive sentence is a sentence we would like to be closer to the anchor sentence in embedding space, and the negative sentence is a sentence we would like to be farther apart from the anchor in embedding space. For example, to fine-tune the model to help it distinguish negation sentences, our training set can be

    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:摘录中关于三元组标签表达任务相关性的定义、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:anchor, positive, negative) examples. For a given anchor sentence, the positive sentence is a sentence we would like to be closer to the anchor sentence in embedding space, and the negative sentence is a sentence we would like to be farther apart from the anchor in embedding space. For example, to fine-tune the model to help it distinguish negation sentences, our training set can be