使用 AWS SageMaker、Greengrass 和 Bedrock 实现实际解​​决方案中的边缘人工智能（由高通赞助）.pdf

1、1 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 2 2 0 2-SEdge AI in Real-World Solutions using AWS SageMaker,Greengrass and BedrockKrishna SridharVP EngineeringQualcomm Technologies,Inc.Ashvin RohariaSenior EngineerQualcomm Technologies,Inc.34On-deviceCloudReal-time Infere

2、nce Is it more efficient to run models on-device or in the cloud?Within 5 minutes,with 5 lines of codeReal-time Intelligent Industrial SystemsHow can we deploy on-device on thousands of devices?Qualcomm AI Engine Direct6Model training with SageMakerModel optimization and deploymentEdge inference on

3、deviceDevice management and analyticsWhat You Will Learn7Meet The StackModel training On-device optimizationML deployment pipelinesDevice management and AnalyticsEdge Inference8OptimizePrepareDeployRunMonitorTrainWorkflow Overview9 Model training environment Enterprise security and compliance Bring

4、your own data Managed training jobs Outputs for Qualcomm AI Hub(ONNX,PyTorch)Step 1:Train model in Amazon SageMakerTrain10Step 2:Optimize for on-device with Qualcomm AI HubTrained ModelPick a devicePick a runtimeCloud TrainingFinetuneBYODTrainQualcomm AI HubOptimized ModelOptimizeTrain11Step 3:Prepa

5、re Deployment in Edge ImpulseAdvantech AOM-DK2721(QCS6490)Upload trained models to Edge Impulse using BYOMOptimizePrepareTrainObtained ready to deploy artifacts on an edge device12Step 4:Deployment with AWS IoT Greengrass3.Deploy ComponentsSetup AWS IOT Greengrass on edge devicesOptimizePrepareDeplo

6、yTrainRegister Edge Impulse components with customized YAML recipesDeploy components to fleet of devices13Step 5:Run Models on Fleetedge-impulse-linux-runnerAdvantech AOM-DK2721(QCS6490)OptimizePrepareDeployRunMonitorTrain14DeployDataCommandsAWS IoT CoreGreeng