Deploy Hugging Face models on Amazon SageMaker AI with coding agents
aws ·
This AWS post explains how coding agents can deploy Hugging Face models on Amazon SageMaker AI using six open-source skills from the Hugging Face Skills repository. Without those skills, agents such as Kiro and Claude Code selected an incompatible or outdated Text Generation Inference container for Qwen3, causing failed deployments and unnecessary GPU charges. The skills guide agents through AWS context discovery, supported Python environment setup, IAM role verification, serving-image selection from the AWS Deep Learning Containers catalog, and deployment with production defaults. These defaults include autoscaling, three Amazon CloudWatch alarms, smoke testing, logging, and a verified teardown process. The example deploys Qwen/Qwen3-0.6B to a real-time endpoint, while the skills also support scale-to-zero, serverless, asynchronous, batch, and Amazon Bedrock Custom Model Import deployments. The skills use Python and the AWS CLI, work on macOS, Linux, and Windows, and require supported Python and AWS account permissions. The post also notes that real-time endpoints continue to incur instance charges until they are deleted.
aws