Amazon Nova recipes

You can get an Amazon Nova recipe from the SageMaker HyperPod recipes repository. An Amazon Nova recipe is a YAML configuration file that provides details to SageMaker on how to run your model customization job. It provides the base model name, sets training hyperparameters, defines optimization settings, and includes any additional options required to fine-tune or train the model successfully.

You can also access Amazon Nova recipes through SageMaker AI Monarch and JumpStart by navigating to the SageMaker JumpStart model hub, selecting AWS, and browsing Amazon Nova models to find their associated recipes. Both SageMaker AI Monarch and JumpStart provide sample notebooks for each recipe, which include all the necessary steps to modify the recipes and run customization jobs using SageMaker training jobs or SageMaker HyperPod environments.

To access the recipes pages in Amazon SageMaker Studio, the execution role must have the following permissions.

JSON

{
    "Version":"2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "s3:GetObject"
            ],
            "Resource": [
                "arn:aws:s3:::*model-customization-recipes*"
            ]
        }
    ]
}

To execute the sample notebooks on SageMaker training jobs and SageMaker HyperPod, use one of the following SageMaker distribution image versions: 2.7.1+, 2.8.0+, 3.2.1+, 3.3.0+. This applies to both SageMaker AI Monarch and JumpStart.

Getting Amazon Nova recipes

To get a base Amazon Nova recipe, clone the SageMaker HyperPod recipes repository by running the following command.

git clone https://github.com/aws/sagemaker-hyperpod-recipes.git

The base recipes are available at recipes_collection/recipes/.

cd recipes_collection/recipes/

The Amazon Nova customization recipes are in the following folders.

Getting Amazon Nova Forge recipes

To access specialized Amazon Nova Forge recipes for SageMaker HyperPod jobs, please set up your access to Amazon Nova Forge and then follow this workshop to set up the Forge version of the SageMaker HyperPod CLI. For SageMaker Training Jobs, the AWS Console and SageMaker Python SDK will automatically access Forge recipes.

Available models and algorithms

The following table summarizes the availability of customization for Amazon Nova 2.0 models and supported algorithms using SageMaker. For Amazon Nova 1.0 model customization, see the Amazon Nova 1.0 recipes.

Amazon Nova 2.0 Lite

The table below lists detailed information of the Amazon Nova 2.0 Lite recipes reference. Scroll horizontally to view the full table, including Image URIs and compute instance types.