Amazon SageMaker SDK adds AI inference optimization to notebooks

Optimizing generative AI inference deployment traditionally involves manual benchmarking across multiple instance types, container versions, and concurrency settings—a time-consuming process of trial and error. The Amazon SageMaker Python SDK integration addresses this friction by automating the entire workflow: the service deploys a model on each candidate configuration, runs a load test matching the user's traffic pattern, and returns a ranked list optimized for the chosen performance target (such as minimizing time-to-first-token for latency-sensitive applications or maximizing throughput for batch workloads). By exposing these capabilities as native SDK operations within the notebook, AWS removes the need to switch to Amazon SageMaker Studio or construct low-level Boto3 API calls, making the optimization workflow fit naturally into existing data science and MLOps pipelines.

The integration also supports production deployment patterns. Data scientists and MLOps teams often operate in separate workflows—the former experimenting and generating recommendations, the latter deploying and monitoring in production. The new `ModelBuilder.from_recommendation_job()` method enables this separation, allowing a completed recommendation job to be hydrated and deployed in a different session or process, thereby bridging experimentation and release cycles.