知识卡片

普通读书笔记卡 · 7010.p · 生成模型:DeepSeek-V4.1-Flash

先按数据量与速度选架构

理解等级:中级

面对四类架构,作者承认选型可能很复杂:模型生命周期、格式差异是一个维度,庞大的实现图景是另一个。但他给出明确推荐:先考虑应用所需的数据量与数据速度——若极低延迟是首要目标,用离线/内存服务(预计算把预测变成查表,最快);否则用 MaaS(最灵活、可扩展、约束最少);若本就运行在边缘设备上,自然用边缘服务。本章余下因此聚焦 MaaS,既因它更灵活,也因它约束更少、更适合作为讲解载体。这个顺序本身是一种工程默认值:把「默认走 MaaS」作为基线,只在延迟极端或设备受限这两个明确条件下偏离,可以避免团队在每种新模型上线时重新做一遍无依据的架构辩论。

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

    • Serving · 查看原书依据

      Our recommended approach is to first consider the amount of data and speed of the data required for your application: if extremely low latency is the priority, use offline/in-memory serving. Otherwise, use MaaS, except when you’re running on an edge device, in which case serving at the edge is (obviously) the most appropriate.

    参考来源

    - 位置:《Reliable Machine Learning》第08章《Serving》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch08.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:Our recommended approach is to first consider the amount of data and speed of the data required for your application: if extremely low latency is the priority, use offline/in-memory serving. Otherwise, use MaaS, except when you’re running on an edge device, in which case serving at the edge is (obviously) the most appropriate.