Deepgram deepens Amazon SageMaker AI observability with Enhanced Metrics
aws ·
Deepgram has introduced two observability features for its speech-to-text and text-to-speech models deployed on Amazon SageMaker AI. The first, Deepgram Enhanced Metrics, publishes billing and usage metrics directly to Amazon CloudWatch without requiring agents or extra permissions, using the existing container logging path. These metrics include ConsumedUnits, which match AWS Marketplace metered billing values, along with audio duration and character counts, enabling customers to reconcile bills and understand feature usage. The second feature adds Prometheus and OpenTelemetry support, which exposes engine-level metrics, per-GPU accelerator metrics, and host metrics through SageMaker AI's detailed observability, queryable via PromQL. This allows customers to monitor GPU utilization, active requests, and engine capacity estimates, with resource labels for filtering by endpoint or instance. Both features operate within AWS Marketplace network isolation constraints, as they use existing logging or host-level collection mechanisms. The article provides examples of querying these metrics using AWS CLI and PromQL, and notes that detailed observability is enabled by default for new endpoints.
aws