Asserting Amazon Nova customization in Amazon SageMaker AI

As we speak, we’re saying a collection of customization capabilities for Amazon Nova in Amazon SageMaker AI. Clients can now customise Nova Micro, Nova Lite, and Nova Professional throughout the mannequin coaching lifecycle, together with pre-training, supervised fine-tuning, and alignment. These methods can be found as ready-to-use Amazon SageMaker recipes with seamless deployment to Amazon Bedrock, supporting each on-demand and provisioned throughput inference.

Amazon Nova basis fashions energy various generative AI use instances throughout industries. As clients scale deployments, they want fashions that replicate proprietary data, workflows, and model necessities. Immediate optimization and retrieval-augmented technology (RAG) work nicely for integrating general-purpose basis fashions into purposes, nevertheless business-critical workflows require mannequin customization to satisfy particular accuracy, price, and latency necessities.

Selecting the best customization approach
Amazon Nova fashions assist a spread of customization methods together with: 1) supervised fine-tuning, 2) alignment, 3) continued pre-training, and 4) data distillation. The optimum selection depends upon objectives, use case complexity, and the supply of information and compute assets. You may as well mix a number of methods to realize your required outcomes with the popular mixture of efficiency, price, and suppleness.

Supervised fine-tuning (SFT) customizes mannequin parameters utilizing a coaching dataset of input-output pairs particular to your goal duties and domains. Select from the next two implementation approaches based mostly on knowledge quantity and price concerns:

• Parameter-efficient fine-tuning (PEFT) — updates solely a subset of mannequin parameters by means of light-weight adapter layers similar to LoRA (Low-Rank Adaptation). It gives quicker coaching and decrease compute prices in comparison with full fine-tuning. PEFT-adapted Nova fashions are imported to Amazon Bedrock and invoked utilizing on-demand inference.
• Full fine-tuning (FFT) — updates all of the parameters of the mannequin and is good for situations when you will have in depth coaching datasets (tens of hundreds of information). Nova fashions personalized by means of FFT may also be imported to Amazon Bedrock and invoked for inference with provisioned throughput.

Alignment steers the mannequin output in direction of desired preferences for product-specific wants and conduct, similar to firm model and buyer expertise necessities. These preferences could also be encoded in a number of methods, together with empirical examples and insurance policies. Nova fashions assist two desire alignment methods:

• Direct desire optimization (DPO) — gives a simple method to tune mannequin outputs utilizing most popular/not most popular response pairs. DPO learns from comparative preferences to optimize outputs for subjective necessities similar to tone and elegance. DPO gives each a parameter-efficient model and a full-model replace model. The parameter-efficient model helps on-demand inference.
• Proximal coverage optimization (PPO) — makes use of reinforcement studying to boost mannequin conduct by optimizing for desired rewards similar to helpfulness, security, or engagement. A reward mannequin guides optimization by scoring outputs, serving to the mannequin be taught efficient behaviors whereas sustaining beforehand discovered capabilities.

Continued pre-training (CPT) expands foundational mannequin data by means of self-supervised studying on massive portions of unlabeled proprietary knowledge, together with inside paperwork, transcripts, and business-specific content material. CPT adopted by SFT and alignment by means of DPO or PPO gives a complete method to customise Nova fashions on your purposes.

Data distillation transfers data from a bigger “instructor” mannequin to a smaller, quicker, and extra cost-efficient “pupil” mannequin. Distillation is helpful in situations the place clients wouldn’t have ample reference input-output samples and might leverage a extra highly effective mannequin to reinforce the coaching knowledge. This course of creates a personalized mannequin of teacher-level accuracy for particular use instances and student-level cost-effectiveness and pace.

Here’s a desk summarizing the accessible customization methods throughout completely different modalities and deployment choices. Every approach gives particular coaching and inference capabilities relying in your implementation necessities.

Early entry clients, together with Cosine AI, Massachusetts Institute of Expertise (MIT) Laptop Science and Synthetic Intelligence Laboratory (CSAIL), Volkswagen, Amazon Buyer Service, and Amazon Catalog Methods Service, are already efficiently utilizing Amazon Nova customization capabilities.

Customizing Nova fashions in motion
The next walks you thru an instance of customizing the Nova Micro mannequin utilizing direct desire optimization on an current desire dataset. To do that, you should utilize Amazon SageMaker Studio.

Launch your SageMaker Studio within the Amazon SageMaker AI console and select JumpStart, a machine studying (ML) hub with basis fashions, built-in algorithms, and pre-built ML options which you could deploy with a number of clicks.

Then, select Nova Micro, a text-only mannequin that delivers the bottom latency responses on the lowest price per inference among the many Nova mannequin household, after which select Prepare.

Subsequent, you’ll be able to select a fine-tuning recipe to coach the mannequin with labeled knowledge to boost efficiency on particular duties and align with desired behaviors. Selecting the Direct Desire Optimization gives a simple method to tune mannequin outputs along with your preferences.

If you select Open pattern pocket book, you will have two setting choices to run the recipe: both on the SageMaker coaching jobs or SageMaker Hyperpod:

Select Run recipe on SageMaker coaching jobs whenever you don’t must create a cluster and prepare the mannequin with the pattern pocket book by deciding on your JupyterLab area.

Alternately, if you wish to have a persistent cluster setting optimized for iterative coaching processes, select Run recipe on SageMaker HyperPod. You may select a HyperPod EKS cluster with no less than one restricted occasion group (RIG) to supply a specialised remoted setting, which is required for such Nova mannequin coaching. Then, select your JupyterLabSpace and Open pattern pocket book.

This pocket book gives an end-to-end walkthrough for making a SageMaker HyperPod job utilizing a SageMaker Nova mannequin with a recipe and deploying it for inference. With the assistance of a SageMaker HyperPod recipe, you’ll be able to streamline complicated configurations and seamlessly combine datasets for optimized coaching jobs.

In SageMaker Studio, you’ll be able to see that your SageMaker HyperPod job has been efficiently created and you’ll monitor it for additional progress.

After your job completes, you should utilize a benchmark recipe to judge if the personalized mannequin performs higher on agentic duties.

For complete documentation and extra instance implementations, go to the SageMaker HyperPod recipes repository on GitHub. We proceed to broaden the recipes based mostly on buyer suggestions and rising ML traits, making certain you will have the instruments wanted for profitable AI mannequin customization.

Availability and getting began
Recipes for Amazon Nova on Amazon SageMaker AI can be found in US East (N. Virginia). Study extra about this characteristic by visiting the Amazon Nova customization webpage and Amazon Nova consumer information and get began within the Amazon SageMaker AI console.

–Betty

Up to date on July 16, 2025 – Revised the desk knowledge and console screenshot.