New serverless customization in Amazon SageMaker AI accelerates model fine-tuning

Amazon SageMaker AI introduces serverless customization to streamline fine-tuning of popular AI models including Amazon Nova, DeepSeek, Llama, and Qwen.

• •Supports four fine-tuning techniques: Supervised Fine-Tuning, Direct Preference Optimization, Reinforcement Learning from Verifiable Rewards (RLVR), and Reinforcement Learning from AI Feedback (RLAIF)
• •Serverless mode automatically provisions compute resources based on model and data size, eliminating infrastructure management
• •UI-based workflow covers model selection, hyperparameter configuration, training job submission, evaluation, and deployment in a few clicks
• •Includes a serverless MLflow integration for automatic experiment metric logging and rich visualizations without code changes
• •Trained models can be deployed to Amazon Bedrock for serverless inference or to SageMaker AI endpoints for controlled instance-level deployment