Machine Learning
Amazon SageMaker Model Monitor
Amazon SageMaker Model Monitor schedules checks for data quality, model quality, bias drift, and feature-attribution drift on supported endpoints. GenAI teams can combine these signals with custom quality and safety telemetry.
Key points
- Baselines define expected constraints and statistics
- Monitoring schedules analyze captured inference data
- Violations are published for alerting and investigation rather than automatically fixing a model
When to use it
- Detect input drift on a custom embedding endpoint
- Alert when a classifier used in an agent workflow departs from its baseline
Exam tips
- Choose Model Monitor for SageMaker endpoint monitoring and CloudWatch for platform-wide operational telemetry
- Capture data selectively and protect it because inference payloads may contain sensitive text