知识卡片
查询似然按候选文档条件计算
QLM对每个候选文档求同一查询token序列的条件似然并据此排序。必须正确对齐目标token并只统计查询部分,logit需变成条件概率;高可生成性并不自动等于事实支持。
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
- · 检索增强生成
原书出处 · 1 处
- · 查看原书依据
Query likelihood model (QLM) A QLM estimates the probability of generating the query given a candidate document as input. You can treat an LLM as a QLM, utilizing its zero-shot capabilities to rank candidate documents based on the query token probabilities. Alternatively, you can fine-tune an LLM on query generation tasks to improve its suitability as a QLM. A typical prompt for a QLM would
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:摘录中关于查询似然按候选文档条件计算的定义、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:Query likelihood model (QLM) A QLM estimates the probability of generating the query given a candidate document as input. You can treat an LLM as a QLM, utilizing its zero-shot capabilities to rank candidate documents based on the query token probabilities. Alternatively, you can fine-tune an LLM on query generation tasks to improve its suitability as a QLM. A typical prompt for a QLM would