← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8708.a · 生成模型:GPT-6

合法标识约束不能保证相关

理解等级:高级

将解码限制在已存在docID集合,可防止生成无法定位的编号。它只保证结果在库中,不保证相关或最新;语料增删时还需更新映射与约束,训练式索引可能需要重新训练。

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

      During inference, the model is provided with a query similar to the setup in Figure 12-3 and asked to generate the correct docID(s) that are relevant to the query. Constrained beam search is used to ensure that the docID generated by the model corresponds to a valid docID in the corpus.

    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:摘录中关于合法标识约束不能保证相关的定义、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:During inference, the model is provided with a query similar to the setup in Figure 12-3 and asked to generate the correct docID(s) that are relevant to the query. Constrained beam search is used to ensure that the docID generated by the model corresponds to a valid docID in the corpus.