AWS
5 min read

Introducing Amazon Nova Forge: Build your own frontier models using Nova

Read Full Article

Summary

The article introduces Amazon Nova Forge, a service designed to help organizations build their own frontier models using generative AI. It addresses the limitations of existing customization approaches by allowing users to start model development from various training checkpoints and blend proprietary data with Amazon Nova-curated datasets. This approach aims to reduce catastrophic forgetting while preserving foundational skills. Nova Forge also integrates with AWS services, providing a seamless experience for users looking to develop AI models tailored to specific industry needs.

Key Learnings

  • 1Amazon Nova Forge allows for the blending of proprietary and curated datasets to enhance model training and reduce catastrophic forgetting.
  • 2The service supports multiple training phases (pre-training, mid-training, post-training) to optimize model development.
  • 3Reinforcement Fine-Tuning (RFT) can be utilized to improve factual accuracy and mitigate hallucinations in specific domains.
  • 4Integration with Amazon SageMaker AI and Amazon Bedrock ensures security and consistent APIs for model deployment.
  • 5The built-in responsible AI toolkit allows customization of safety and content moderation settings according to business requirements.

Who Should Read This

Senior AI Engineers developing customized generative AI models for industry-specific applications

Test Your Knowledge

?

What are the trade-offs between using early checkpoints versus fully trained models in the context of Nova Forge?

?

How does the data mixing approach in Nova Forge help mitigate catastrophic forgetting during model training?

?

What specific scenarios might lead to the failure of a model when using Continued Pre-Training (CPT) with proprietary data?

?

In what ways can organizations leverage the reinforcement learning capabilities of Nova Forge to enhance model performance?

?

Why is it important to configure safety and content moderation settings in models developed with Nova Forge?

?

What considerations should be made when selecting datasets to blend with Amazon Nova-curated data?

Topics

Read Full Article at AWS

More articles about Amazon Bedrock

Explore Amazon Bedrock engineering →
AWS
5m

AWS Weekly Roundup: Amazon Connect Health, Bedrock AgentCore Policy, GameDay Europe, and more (March 9, 2026)

The article provides a comprehensive overview of recent updates and launches from AWS, highlighting innovations such as Amazon Connect Health, which offers AI-driven solutions for healthcare, and the...

AWS
4m

Introducing OpenClaw on Amazon Lightsail to run your autonomous private AI agents

The article introduces OpenClaw, an autonomous private AI agent, now available on Amazon Lightsail. It details the process of launching an OpenClaw instance, which is pre-configured with Amazon...

AWS
6m

AWS Weekly Roundup: OpenAI partnership, AWS Elemental Inference, Strands Labs, and more (March 2, 2026)

The article provides an overview of the latest developments from AWS, including a strategic partnership with OpenAI aimed at enhancing AI capabilities for enterprises. It highlights the introduction...

AWS
5m

AWS Weekly Roundup: Claude Sonnet 4.6 in Amazon Bedrock, Kiro in GovCloud Regions, new Agent Plugins, and more (February 23, 2026)

The AWS Weekly Roundup highlights significant updates in AI and cloud services, including the introduction of Claude Sonnet 4.6 in Amazon Bedrock, which enhances coding and professional work...

AWS
6m

AWS Weekly Roundup: Amazon EC2 M8azn instances, new open weights models in Amazon Bedrock, and more (February 16, 2026)

The AWS Weekly Roundup highlights significant updates including the launch of Amazon EC2 M8azn instances, which are powered by fifth generation AMD EPYC processors, offering enhanced performance...

More from AWS Engineering

View AWS engineering blogs →