brunopistone / ml-edge-getting-started Goto Github PK

This repository contains examples of different models, showing how they can be built using SageMaker and prepared for deployment at the edge.

In this repository, we will use Amazon Sagemaker to train the models, Sagemker-Neo to compile them Sagemaker Edge Manager to packaged them. Once the package is created, it can be deployed at the edge devices(for example: Jetson Xavier Jetpack 4.4.1)

Let us understand the structure of the repository:

• models folder contains one sub-folder per model/framework type. These subfolders contain a jupyter notebook which does the following:

  • Download the model from the official repository
  • Train the model on a sample dataset
  • Compile the model for an edge device(in this case, a Jetson Xavier Jetpack 4.4.1)
  • Create a SageMaker Edge Manager Packaging job to prepare the deployment package

• sagemaker_edge_manager_agent_docker folder contains the details on how to build a docker container for SageMaker Edge Agent.

Below is the list of models with their destination hardware platform, processor and other details.

Instructions of how to create a docker container for the ARM64 agent + Nvidia devices -> Nano/Xavier.

See CONTRIBUTING for more information.

This library is licensed under the MIT-0 License. See the LICENSE file.