Bring your own model with Amazon SageMaker AI: Script mode in SDK v3
aws ·
The AWS blog post explains how to use script mode with Amazon SageMaker AI SDK v3 to bring your own model. The new SDK unifies framework-specific estimators into a single ModelTrainer for training and ModelBuilder for deployment. A key feature is the SourceCode object, which syncs a local code directory into the container at runtime, eliminating the need to rebuild Docker images for each code change. The post describes two end-to-end examples: training and deploying a scikit-learn Random Forest on the diabetes dataset, and fine-tuning Stable Diffusion 3.5 with LoRA. It provides detailed steps for building a minimal container, configuring source code and compute resources, launching a training job, and deploying a real-time endpoint with DJL Serving. It also compares SDK v2 and v3 and lists prerequisites such as an AWS account, IAM role, S3 bucket, and Docker access.
aws