Amazon SageMaker AI Model and Endpoint Deployment Pipeline with AWS Step Functions

Description

Simplify your machine learning workflow with this AWS Step Functions State Machine. Deploy SageMaker AI Models and Endpoints effortlessly, without the need for SageMaker AI Studio. Key features include:

Version control: Preserve your model history using SageMaker AI Model Cards

Flexibility: Support for multi-container SageMaker AI Models and dependent ML models (also known as inference pipelines), as well as real-time and asynchronous SageMaker AI Endpoints

Streamlined invocation: Leverage Systems Manager Parameters to call your SageMaker AI models in the correct order for your Inference Pipeline

This solution empowers developers to manage their entire ML deployment process through a single, automated workflow. By eliminating manual steps and providing a consistent deployment pattern, you can accelerate your path from experimentation to production.

Visuals

• Model Deployment Task: Deploys SageMaker AI Models and Model Cards
• Endpoint Deployment Task: Deploys SageMaker AI Endpoints and Endpoint Configurations
• Endpoint Scaling and SSM Task: Updates SageMaker AI Endpoint Scaling, and creates SSM Parameters for the Endpoints
• Update Model DAG Task: Updates the Model DAG (adds or removes parameters from SSM)

Installation

1. Clone the repo and ensure the correct packages are installed
2. An AWS Account is required
3. Update the Model Containers JSON File in state_machine_input/model_containers.json.
   • This will store all of the information about your SageMaker AI Models and SageMaker AI Endpoints
4. Add your Model Card JSON Files to the model_card_json_files folder

Usage

1. Deploy CDK Resources:
   • Begin by deploying the AWS CDK (Cloud Development Kit) resources to your AWS account. This sets up the necessary infrastructure for the subsequent steps.
   • Configure your AWS Account in the CDK Stack, or through environment variables
   • In the root directory of this project, run cdk synth followed by cdk deploy
2. Run Model Deployment State Machine:
   • After deploying the CDK resources, execute the Model Deployment Step Functions State Machine. This will create the SageMaker AI Models and endpoints for you.
3. Run Inference:
   • Once the model deployment is complete, you can start running inference using the newly created SageMaker AI endpoints.

Support

For help, reference this APG or use AWS Documentation here

Authors and acknowledgment

• Anna Heltz, heltzan@amazon.com
• Jack Tanny, johtanny@amazon.com

Security

See CONTRIBUTING for more information.

License

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