Amazon SageMaker AI
Amazon SageMaker AI is a fully managed machine-learning service from Amazon Web Services for preparing data and building, training, deploying, and governing ML and foundation models with managed infrastructure, tools, and workflows.
AI Practitioner focus
- The service was renamed Amazon SageMaker AI on December 3, 2024; API, CLI, IAM, endpoint, and CloudFormation names retain
SageMakerfor compatibility. - It covers custom ML preparation, training, tuning, deployment, MLOps, monitoring, Clarify, Model Cards, Feature Store, Data Wrangler, and Amazon SageMaker JumpStart.
- Choose SageMaker AI for control over custom/open models and the ML lifecycle; choose Amazon Bedrock for managed access to supported FMs and GenAI building blocks with less infrastructure management.
Key points
- SageMaker Studio is the current web experience for ML workflows; Studio Classic is no longer open for onboarding new applications.
- Supports built-in algorithms and frameworks plus bring-your-own code/containers for distributed training and managed hosting.
- SageMaker Data Wrangler prepares data, Feature Store manages reusable features, Clarify analyzes bias/explainability, Model Monitor detects production quality/drift issues, and Model Cards document governance details.
- Amazon SageMaker JumpStart supplies pretrained foundation/task-specific models and algorithms that can be evaluated, customized, and deployed to SageMaker endpoints.
- Distinct from managed task APIs such as Amazon Rekognition, Amazon Comprehend, and Amazon Polly, which solve predefined problems without owning a custom ML lifecycle.
Pricing
- Pay for the compute, storage, data processing, training, inference, and other components used; pricing and minimum billing units vary by feature.