Amazon Nova just unlocked its most comprehensive suite of customization options yet. Now, you can precisely tailor Nova models — Micro, Lite, Pro, and Canvas — across all stages of the AI lifecycle to match proprietary data, workflows, and branding. Whether your goal is rapid iteration or tackling highly specialized tasks, you now have a Nova customization path for every scenario via Amazon Bedrock and SageMaker AI.

Customization Options & Use Cases

Supervised Fine-Tuning (SFT)

Improves model accuracy for specific tasks using your labeled data.

1. Parameter-Efficient Fine-Tuning (PEFT) — Best for fast, affordable tuning with limited compute
2. Full Fine-Tuning — Best for maximum accuracy with large datasets

Alignment

Tunes model output for brand voice, compliance, or experience.

1. Direct Preference Optimization (DPO) — Adjusting outputs with preference data
2. Proximal Policy Optimization (PPO) — Reinforcement learning for helpfulness, safety, or engagement

Continued Pre-Training (CPT)

Expands Nova’s knowledge with large, proprietary, unlabeled datasets.

Knowledge Distillation

Copies intelligence from a large “teacher” model to a smaller, cheaper “student” model.

Learn more