知识卡片

普通读书笔记卡 · 8708.b · 生成模型:GPT-6

紧耦合检索优化下游收益

理解等级:高级

检索器可从生成模型回答质量或正确答案似然学习,取回更能帮助该模型的证据。与通用相关性训练目标不同,收益可能依赖目标模型版本;更换生成器后应重新检验迁移。

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

      Next, let’s discuss tightly-coupled retrievers, another new topic in the retrieval space. Tightly-coupled retrievers As seen in Chapter 11, in embedding-based retrieval, the embedding model is typically independent of the language model to which the retrieval results are fed. We will refer to them as loosely-coupled retrievers. In contrast, a tightly-coupled retriever is trained such that it learns from LLM feedback; the model learns to

    参考来源

    - 位置:《Designing Large Language Model Applications》第12章《检索增强生成》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch12.html) - 结论依据:摘录中关于紧耦合检索优化下游收益的定义、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:Next, let’s discuss tightly-coupled retrievers, another new topic in the retrieval space. Tightly-coupled retrievers As seen in Chapter 11, in embedding-based retrieval, the embedding model is typically independent of the language model to which the retrieval results are fed. We will refer to them as loosely-coupled retrievers. In contrast, a tightly-coupled retriever is trained such that it learns from LLM feedback; the model learns to