← 返回刚才阅读的位置

知识卡片

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

MaaS:以 API 提供模型能力

理解等级:中级

模型即服务(MaaS)类似 SaaS,天生偏向微服务架构:模型存放在专用集群、通过定义良好的 API(传输与序列化用 gRPC 或 REST 皆可)提供结果,因以微服务形态提供而与主应用相对隔离,不必进程内交互或紧耦合,因此是最灵活、最可扩展的部署/服务策略。优势成组:可用实时上下文与新特征做实时预测;关注点分离让 ML 工程更稳地调整模型并套用成熟运维手法,多数 MaaS 模型可做成无状态无共享配置依赖,加容量只是加实例(水平扩展);部署隔离让应用与模型服务各选技术栈、按不同节奏发布到 QA/预发/金丝雀;版本管理易扩散(同集群存多版本按需指向,极便于 A/B 测试,还可把模型标识设计进响应以跟踪路由与归集事件数据,支持滚动重部署);集中化使系统健康、容量/吞吐、延迟、资源消耗与每模型展示/点击/转化等业务指标更易监控,把输入输出包一层并标准化模型识别加载后,许多 SRE「黄金四信号」类指标可接现成工具顺带获得。

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

      Model as a Service The Model-as-a-Service (MaaS) approach is similar to software as a service and inherently favors a microservice architecture. With MaaS, models are stored in a dedicated cluster and served results via well-defined APIs. Regardless of the transport or serialization methods (e.g., gRPC or REST),6 because models are served as a microservice, they’re relatively isolated from the main application (Figure 8-4).

    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:Model as a Service The Model-as-a-Service (MaaS) approach is similar to software as a service and inherently favors a microservice architecture. With MaaS, models are stored in a dedicated cluster and served results via well-defined APIs. Regardless of the transport or serialization methods (e.g., gRPC or REST),6 because models are served as a microservice, they’re relatively isolated from the main application (Figure 8-4).