Nova Forge SDK series part 2: Practical guide to fine-tune Nova models using data mixing capabilities

2026-04-17 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Software Development & Engineering · Depth: Advanced, short

Summary

This guide details the process of fine-tuning an Amazon Nova model using the Amazon Nova Forge SDK, focusing on data mixing techniques. It outlines a five-stage workflow: environment setup, data preparation, training configuration, model training, and model evaluation. The article emphasizes data mixing to fine-tune models on domain-specific data without compromising general capabilities, citing a previous demonstration where this approach preserved near-baseline Massive Multitask Language Understanding (MMLU) scores while achieving a 12-point F1 improvement on a Voice of Customer classification task with 1,420 categories. Prerequisites include an AWS account with Nova Forge access, a SageMaker HyperPod cluster (using `ml.p5.48xlarge` instances), SageMaker MLflow for tracking, and an IAM role with necessary permissions. The guide uses the MedReason dataset for a medical reasoning use case.

Key takeaway

For ML Engineers customizing large language models for enterprise applications, this guide provides a repeatable playbook for fine-tuning Amazon Nova models with data mixing. Your team should adopt this data mixing strategy to enhance domain-specific performance without sacrificing the model's broader intelligence. Consider starting with short test runs to validate configurations and manage costs associated with high-end GPU instances.

Key insights

Fine-tuning Amazon Nova models with data mixing preserves general capabilities while improving domain-specific performance.

Principles

• Data mixing prevents loss of general model capabilities.
• Token-level validation is crucial for training data integrity.

Method

The workflow involves installing the Nova Forge SDK, configuring AWS resources, preparing and sanitizing training data, configuring SageMaker HyperPod and MLflow, launching a LoRA-based supervised fine-tuning job, and evaluating the model.

In practice

• Sanitize training data to avoid conflicts with model chat templates.
• Use SageMaker HyperPod for distributed GPU training.
• Track experiments with SageMaker MLflow.

Topics

• Amazon Nova Forge SDK
• Nova Model Fine-tuning
• Data Mixing
• SageMaker HyperPod
• Low-Rank Adaptation

Code references

• aws-samples/amazon-nova-samples
• aws/nova-forge-sdk

Best for: Machine Learning Engineer, AI Engineer, MLOps Engineer

Related on AIssential

• Scale LLM fine-tuning with Hugging Face and Amazon SageMaker AI — Specialized LLMs fine-tuned on proprietary data offer enterprises accuracy, security, and domain-specific knowledge.
• Harnessing non-adversarial robustness in large language models — LLM robustness to prompt variations can be efficiently acquired via fine-tuning, avoiding full retraining.
• The art and science of hyperparameter optimization on Amazon Nova Forge — Customizing LLMs on Amazon Nova Forge requires balancing strategic choices and systematic hyperparameter tuning to prevent catastrophic forgetting.
• Is Fine-Tuning Still Valuable? — Effective LLM fine-tuning requires robust evaluation systems and iterative data curation, especially for domain-specific tasks.
• Transitioning from Azure Language Features to Foundry Models — Azure AI Language features are deprecating, requiring migration to Foundry models using LLMs, RAG, or fine-tuning for equivalent functionality.
• Fine-Tuning Qwen3.5 — Fine-tuning small vision-language models like Qwen3.5-0.8B on domain-specific datasets significantly improves specialized task performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.