知识卡片
乘积量化用子空间码本压缩向量
将向量拆为子向量,各自以训练的码本代表并存码字,查询可据码本近似距离。它与简单逐坐标转整数不同,压缩会引入误差;需用代表性训练数据、重排与任务评估控制代价,不能保证无损保留全部细微信息。
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
原书出处 · 1 处
- · 查看原书依据
The product quantization you would use to implement semantic search needs to be more complicated than that. The vectors you need to compress are much longer (have many more dimensions), and the compression needs to be much better at retaining all the subtle bits of information in the vectors. This is especially important for plagiarism and LLM detectors. It turns out that you can achieve
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第10章《真实世界中的大语言模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-10.html)
- 结论依据:摘录中关于乘积量化用子空间码本压缩向量的定义、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:The product quantization you would use to implement semantic search needs to be more complicated than that. The vectors you need to compress are much longer (have many more dimensions), and the compression needs to be much better at retaining all the subtle bits of information in the vectors. This is especially important for plagiarism and LLM detectors. It turns out that you can achieve