This document is the API specification for the AWS AI toolkit for DJL.

The aws-ai module contains classes that make it easy for DJL to access AWS services.

You can use this toolkit to load models from AWS S3 bucket directly:

    // Set model zoo search path system property. The value can be
    // comma delimited url string. You can add multiple s3 url.
    // The S3 url should point to a folder in your s3 bucket.
    // In current implementation, DJL will only download files directly
    // in that folder. The archive file like .zip, .tar.gz, .tgz, .tar.z
    // files will be extracted automatically. This is useful for the models
    // that are created by AWS SageMaker.
    // The folder name will be interpreted as artifactId and modelName.
    // If your model file has a different name then the folder name, you
    // need use query string to tell DJL which model you want to load.
    System.setProperty("ai.djl.repository.zoo.location",
            "s3://djl-misc/test/models/resnet18?model_name=resent18_v1");

    Criteria<Image, Classifications> criteria =
        Criteria.builder()
                .optApplication(Application.CV.IMAGE_CLASSIFICATION)
                .setTypes(Image.class, Classifications.class)
                .optArtifactId("ai.djl.localmodelzoo:resnet18")
                .optProgress(new ProgressBar())
                .build();

    ZooModel<Image, Classifications> model = criteria.loadModel();

Packages

Package	Description
ai.djl.aws.s3	Contains a built-in implementation of Repository class that can be used for loading models from AWS S3 bucket directly.
ai.djl.aws.sagemaker	Contains a utility classes to deploy model on Amazon SageMaker.