知识卡片

普通读书笔记卡 · 9393.b · 生成模型:GPT-6

缓存收益依重复率与存储开销

理解等级:中级

用户输入重复较多时可减少推理,输入大多唯一则可能增加哈希、读写和存储成本。测量命中率与延迟并设保留策略,缓存查找也不必线性遍历所有历史;缓存内容可能含用户文本,应按用途控制。

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

      To keep CPU costs down, you can cache the results of each prediction to reuse it without even running your NLP pipeline! And this can help a lot when your users are triggering lots of requests to the servers running your endpoint. A cache stores the mapping between the requests (inputs) and the responses (NLP results) on disk within your server. Then, when another request

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》附录E《附录E:容器化微服务部署》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/appendix-e.html) - 结论依据:摘录中关于缓存收益依重复率与存储开销的定义、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:To keep CPU costs down, you can cache the results of each prediction to reuse it without even running your NLP pipeline! And this can help a lot when your users are triggering lots of requests to the servers running your endpoint. A cache stores the mapping between the requests (inputs) and the responses (NLP results) on disk within your server. Then, when another request