Inside the Amazon Nova Forge
Source: Dev.to
Source: Dev.to – Inside the Amazon Nova Forge
Amazon Nova Forge
Amazon Nova Forge is a development environment inside Amazon SageMaker AI that lets you build Novellas—private, custom versions of Amazon’s Nova frontier models.
Unlike most AI services that only let you use a model or fine‑tune its final layer, Nova Forge introduces Open Training. With Open Training you can:
- Access the model at multiple life‑stage checkpoints.
- Embed your company’s proprietary knowledge directly into the model’s core reasoning, not just its output layer.
This blog post is an introduction to Amazon Nova Forge and what makes its training process unique.
What Makes It Different?
- Prompt engineering and RAG provide external context but don’t change a model’s core intelligence.
- Standard fine‑tuning happens too late in the lifecycle, trying to steer a “finished” model that is already set in its ways.
Nova Forge moves customization earlier—into the training process—so specialized knowledge is embedded where it actually sticks.
Positioning
| Platform | Primary Use | Nova Forge Advantage |
|---|---|---|
| Amazon Bedrock | Consuming models (you can fine‑tune, but you work with a black‑box) | Build the model itself using deeper training techniques |
| Azure AI / Google Vertex AI | Fine‑tuning frontier models (no access to intermediate checkpoints) | Data Blending – mix your data with Amazon’s original training data to prevent forgetting |
Terminology
- Novella – The custom model you create; a “private edition” of Nova.
- Checkpoints – Saved states of the model during its initial training (pre‑training, mid‑training, post‑training).
- Data Blending – Mixing your proprietary data with Nova‑curated datasets so the model stays smart while learning your specific business.
- Reinforcement Fine‑Tuning (RFT) – Using reward functions (logic‑based feedback) to teach the model complex, multi‑step tasks.
- Catastrophic Forgetting – When a model learns new information but loses its original abilities; Nova Forge is designed to prevent this.
The Workflow: From Training to Production
The process bridges the gap between the lab (Amazon SageMaker) and the app (Amazon Bedrock).
- Selection – Choose a Nova base model and a specific checkpoint (e.g., a mid‑training checkpoint) in Amazon SageMaker Studio.
- Training (SageMaker AI) – Use SageMaker Recipes (pre‑configured training scripts) to blend your data from S3 with Nova’s datasets. Compute runs on SageMaker’s managed infrastructure.
- Refinement – (Optional) Run RFT in SageMaker to align the model with specific business outcomes or safety guardrails.
- Deployment (Bedrock) – Import the finished Novella into Amazon Bedrock as a private model.
- Production – Applications call the custom model via the standard Bedrock API, benefitting from Bedrock’s server‑less scaling and security.
Sample Training Workflow
Data Privacy and Protection
The security model is the most critical part of any deployment:
- Sovereignty – Your data remains in your own S3 buckets and stays within your VPC boundaries.
- No Leakage – AWS explicitly states that customer data is not used to train the base Amazon Nova models. Your Novella instance is a private resource visible only to your AWS account.
- Encryption – Data is encrypted at rest with AWS KMS (either AWS‑managed or customer‑managed keys) and in transit with TLS 1.2 or higher.
- Governance – Access is controlled through standard IAM policies, and all training activity is logged in CloudTrail.
Pricing Model
Nova Forge has a distinct cost structure reflecting its “frontier” status:
- Subscription Fee – Access to the Forge environment starts at approximately $100,000 per year.
- Usage Costs – In addition to the subscription, you pay for the SageMaker compute (GPUs) used during the training phase.
- Comparison – Cheaper than training from scratch: Building a frontier model from zero costs millions in compute and months of R&D. Nova Forge provides shortcuts to achieve comparable results for a fraction of that cost.
End of article.
Basic Fine‑Tuning
Standard fine‑tuning on Bedrock is much cheaper (often just a few dollars per hour), but it cannot achieve the deep “domain‑native” intelligence that Nova Forge provides.
Summary
Amazon Nova Forge marks a shift from generic AI to native intelligence, where models don’t just reference your data—they are built from it. By using Open Training, you can bake specialized knowledge into the model’s core at the pre‑training or mid‑training stages. This results in a private Novella that understands your specific industry as naturally as its base language.
Organizations that manage high‑value proprietary data should consider moving beyond treating that information as an external reference. If your workflows involve specialized terminology or regulated processes that standard LLMs struggle to master, shifting customization earlier in the training lifecycle is often more effective than basic fine‑tuning.
Disclaimer: AI tools were used to research and edit this article. Graphics are created using AI.
## Additional References
- [Amazon Nova Forge](https://docs.aws.amazon.com/sagemaker/latest/dg/nova-forge.html)
- [Introducing Amazon Nova Forge: Build your own frontier models using Nova](https://aws.amazon.com/blogs/aws/introducing-amazon-nova-forge-build-your-own-frontier-models-using-nova/)
---
### About the Author