AWS re:Invent 2025 - Amazon Nova Forge: Build your own frontier models using Amazon Nova (AIM3325)

Source: Dev.to

Overview

AWS re:Invent 2025 – Amazon Nova Forge: Build your own frontier models using Amazon Nova (AIM3325)

In this session, Mark Andrews and Karan Bhandarkar introduce Amazon Nova Forge, a service that enables organizations to build custom foundation models using their proprietary data. The session explains how foundation models are constructed through pre‑training, mid‑training, and post‑training phases. Nova Forge provides access to checkpoints at each stage, allowing customers to inject their data at the optimal point while blending it with Amazon Nova’s curated training data to prevent catastrophic forgetting.

Key features

• SageMaker Hyperpod recipes for simplified training
• Support for custom reinforcement‑learning environments
• Configurable safety controls

Customer highlights

• Reddit (Rosa Català) – 26‑point precision improvement and 25 % reduction in missed threats for content moderation.
• Nimbus Therapeutics, Nomura Research Institute, and Sony Group – performance gains in drug discovery, financial services, and legal research while maintaining cost and latency benefits.

This article is auto‑generated from the original presentation content; minor typos or inaccuracies may be present.

Introduction to Nova Forge: Bridging the Gap Between Foundation Models and Organizational Knowledge

Good afternoon, folks. This is the first session of Wednesday afternoon. I’m Mark Andrews, joined by Karan Bhandarkar, principal product manager, and Rosa Català from Reddit.

Agenda

1. How foundation models are constructed – visual overview.
2. Building your frontier model with Nova Forge (led by Karan).
3. Reddit experience – results and insights (Rosa).
4. Additional customer stories (Karan).

Foundation models are incredibly capable, but many organizations have proprietary intellectual property that these models do not know. Traditional Retrieval‑Augmented Generation (RAG) provides a search‑and‑retrieve layer but does not embed the intelligence of your IP directly into the model.

Current Approaches and the Nova Forge Advantage: From RAG to Custom Foundation Models

• RAG – retrieves context for models but leaves the model unchanged.
• LoRA adapters – lightweight fine‑tuning; limited scope.
• Expansion (continued pre‑training) – risk of catastrophic forgetting, which Nova Forge mitigates.

Building a model from scratch is time‑consuming and costly in data acquisition and GPU hours. Nova Forge fast‑tracks development while significantly lowering cost, enabling customers to leverage their IP throughout the training pipeline.

Nova Model Family

• Nova Micro, Lite, Pro – introduced last year.
• Nova Lite – highly capable, general‑purpose model.
• Nova Pro – most capable model (early access).
• Nova Omni – multimodal understanding model (preview).

These models achieve performance comparable to leading frontier models on industry benchmarks, providing a strong foundation for further customization.

Building Foundation Models from Scratch: Pre‑training, Mid‑training, and Post‑training Explained

The process starts with an empty model and proceeds through three stages:

1. Pre‑training – learns generic language patterns from large, diverse datasets.
2. Mid‑training – incorporates domain‑specific data to align the model with organizational knowledge.
3. Post‑training – fine‑tunes the model for specific tasks, safety, and compliance requirements.

Nova Forge supplies checkpoints at each stage, allowing you to inject your proprietary data at the optimal point and blend it with Amazon Nova’s curated data to maintain model stability.

Images & Video Links