从零到生产：使用 Amazon EKS 自动模式简化 AI 部署.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.C N S 3 5 7Zero to Production Simplify AI deployments with Amazon EKS Auto ModeApoorva KulkarniPrin.Specialist SA,ContainersAWSIkenna IzugbokwePrin.Solutions Archite

2、ctAWS 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaOverview of Amazon EKS ML workloads on AWSML Production Deployment ChallengesLeveraging Amazon EKS Auto ModeAmazon EKS Auto Mode for Inference Architecture 2025,Amazon Web Services,Inc.or its affiliates.All rights reserve

3、d.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Overview of EKS ML workloads on AWS 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Customers use millions of GPU-powered instances with Amazon EKS2X+GPU instances with Amazon EKS has more than doubled since 2024

4、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.“Modern professional MLDS developersare expected to have some proficiencywith various aspects of backendinfrastructure”Source:Q1 2025 State of Cloud Native Development for CNCF 2025,Amazon Web Services,Inc.or its affiliates.All righ

5、ts reserved.KubernetesAI/MLKubernetes AI/ML CoreDNSMetrics ServerPrometheusKarpenterEnvoyGrafanaIstioCubeFSCiliumKyvernoHarborSpiffeFluentdJaegerKubecostACKCrossplaneCSICNIPod IdentitySR IOVCert ManagerExternal DNSSecretsNginxArgoCDPyTorchScikit-LearnTensorflowMPIKerasHugging FaceTransformersLangCha

6、inAccelerateLangFuseOpenNNStrandsMXNetCUDAXGBoostCaffeDaskDeepSpeedFlaxHorovodHiveMindJaxPEFTRayTritonAlibiLlaMaKubeflowKuberayDynamovLLMllm-dAI Gateway AIBrix KServe DRANeuron PluginGPU PluginKueueNCCLVolcanoKitOpsMLFlowLeader-Worker SetEFA PluginHPAFeastSeldonSpark OperatorJupyterLab 2025,Amazon W