知识卡片
离线批量推理:把预测变成查表
作者把服务架构归为离线、在线、模型即服务、边缘四大类。离线服务(批量推理)通常最简单最快:面向用户的应用不直接接触模型,模型提前训练好,在离线环境对一组预定义输入跑完预测,把结果存成普通数据集(数据库、CSV 或内存数据结构)。本质是把「按需模型预测」转换成「从存储加载静态数据」这一标准问题——yarnit.ai 可离线算好各商品对某类用户的热度、用作渲染任意条目时的排序辅助,也可把查询意图分类器(具体商品 vs 宽泛品类)结果写进平面文件或内存结构,帮查询引擎改写检索以提升效率。这种做法绕开了「为了随时可被请求而托管模型」的整套复杂性,因此特别适合查询空间有限、对预测新鲜度要求不高的场景。
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
- Serving · 查看原书依据
Offline Serving (Batch Inference) Offline serving is often the simplest and fastest architecture to implement. The application serving the end user is not exposed to the models directly. Models are trained ahead of time, often referred to as batch inference.
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:Offline Serving (Batch Inference) Offline serving is often the simplest and fastest architecture to implement. The application serving the end user is not exposed to the models directly. Models are trained ahead of time, often referred to as batch inference.