AWS updates SageMaker for faster machine learning deployments
Amazon Web Services on Wednesday rolled out a series of updates to SageMaker, the AWS service that helps customers build, train and deploy machine learning models. The new capabilities are designed to make it cheaper and easier to use machine learning.
First, Amazon SageMaker Ground Truth Plus provides customers with access to workforces trained in data labeling. The service makes it faster to create high-quality training datasets, and Amazon says it reduces the costs of creating datasets by 40%. Typically, the process involves building labeling applications and managing a labeling workforce.
Meanwhile, the new SageMaker Training Compiler automatically compiles a user's Python training code and generates GPU kernels specifically for their model. The training code will use less memory and compute, training models faster. AWS says it can speed up training by up to 50%.
Next, Amazon SageMaker Inference Recommender automates load testing and optimizes model performance across machine learning instances. This saves MLOps Engineers the time they'd spend selecting ML instance types and writing custom scripts to run performance benchmarks and load testing.
AWS also rolled out a preview of SageMaker Serverless Inference, a new inference option that lets users deploy ML models for inference without having to configure or manage the underlying infrastructure. The new option in SageMaker automatically provisions, scales, and turns off compute capacity based on the volume of inference requests. Customers pay only for the duration of running the inference code and the amount of data processed.