Machine Learning
Amazon SageMaker AI
Amazon SageMaker AI provides managed capabilities to build, train, tune, deploy, and monitor machine learning models. It is the primary choice when a GenAI team needs custom algorithms, containers, endpoints, accelerators, or detailed deployment control.
Key points
- Real-time, asynchronous, serverless, and batch inference address different traffic patterns
- Endpoint variants enable canary and shadow deployment strategies
- Training and processing jobs use isolated managed compute
When to use it
- Host a fine-tuned open-weight language model
- Deploy a custom reranker with accelerator-specific serving code
Exam tips
- Choose SageMaker AI for custom model lifecycle control and Bedrock for managed foundation-model APIs
- Scale on concurrency, queue depth, memory, and token throughput rather than CPU alone