Amazon SageMaker AI now supports serverless model customization for NVIDIA Nemotron 3.5 Lightning
Amazon SageMaker AI introduces serverless model customization for the NVIDIA Nemotron 3.5 Lightning model, allowing users to tailor it with proprietary data to reduce costs and latency.
Amazon SageMaker AI supports serverless model customization for the NVIDIA Nemotron 3.5 Lightning model, a latest open-weight model with 3B active and 30B total parameters using a hybrid Mixture-of-Experts architecture. This allows deploying and customizing the model on SageMaker AI for specific domains or workflows. Model customization fine-tunes the foundation model on proprietary data, maintaining cutting-edge quality with smaller, cost-effective models that reduce latency. It improves domain-specific task accuracy with SFT, aligns outputs to organizational tone using DPO, and enhances new task performance with RFT. Serverless customization lets SageMaker AI handle all infrastructure provisioning and training orchestration, enabling focus on data and evaluation instead of cluster management, with pay-as-you-go pricing. The feature is available in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Ireland). To get started, initiate a customization job from the Amazon SageMaker Studio model page or programmatically access it using the SageMaker Python SDK. For more information, see the Amazon SageMaker AI model customization documentation.
Why it matters
Amazon SageMaker AI is a platform that simplifies AI model training and deployment, and this update allows users to customize models more efficiently.