Machine Learning
Amazon SageMaker Processing
Amazon SageMaker Processing runs managed data-processing, evaluation, and analysis containers on temporary compute. It is useful for repeatable offline GenAI jobs that need custom code without maintaining a cluster.
Key points
- Processing jobs mount declared inputs and write outputs to configured locations
- Built-in or custom containers define the execution environment
- Instances terminate after the job, so durable outputs must leave local storage
When to use it
- Run a large prompt regression suite before release
- Transform documents and compute offline retrieval metrics
Exam tips
- Choose Processing for bounded ML jobs and EMR for distributed big-data frameworks at larger scale
- Pin the container digest and input versions so evaluation evidence is reproducible