Customizing Models

🆕 V3 EXCLUSIVE FEATURE
Model customization with specialized trainers is available only in SageMaker Python SDK V3, built from the ground up for foundation model fine-tuning. It streamlines the complex process of customizing AI models from months to days with a guided UI and serverless infrastructure that removes operational overhead. Whether you are building legal research applications, customer service chatbots, or domain-specific AI agents, this feature accelerates your path from proof-of-concept to production deployment.

Key Benefits

• Serverless Training: Fully managed compute infrastructure that abstracts away all infrastructure complexity, allowing you to focus purely on model development

• Advanced Customization Techniques: Comprehensive set of methods including supervised fine-tuning (SFT), direct preference optimization (DPO), reinforcement learning with verifiable rewards (RLVR), and reinforcement learning with AI feedback (RLAIF)

• AI Model Customization Assets: Integrated datasets and evaluators for training, refining, and evaluating custom models

• Production Ready: Built-in evaluation, monitoring, and deployment capabilities with automatic resource management

• Open weight model customization
  • Creating assets for model customization
  • AI Model Customization Job Submission
• Amazon Nova Foundation Model
  • Nova Data Mixing
  • Model Customization using SageMaker Training Job
  • Part 2: Model Evaluation
• Model Evaluation
  • SageMaker LLM-as-Judge Evaluation
  • Part 2: LLM-as-Judge with Custom Models
  • SageMaker InspectAI Evaluation
  • SageMaker Custom Scorer Evaluation - Demo
  • SageMaker Benchmark Evaluation - Basic Usage
• Model Deployment
  • Deploy Models to SageMaker Endpoint
  • Deploy Models to Bedrock with BedrockModelBuilder