The Deep Java Library (DJL) is a library developed to help Java developers get started with deep learning. This project is a Spring Boot starter that allows Spring Boot developers to start using DJL for inference.
The starter supports dependency management and auto-configuration.
The released artifacts are available from maven central.
Unreleased artifacts (snapshots) can be downloaded using snapshot repository: https://oss.sonatype.org/conten/repositories/snapshots/.
Dependencies are divided into two sections: starter dependencies, which includes the libraries, and Spring autoconfiguration for Apache MXNet.
The starter configures dependencies using either platform-specific or automatic configuration.
MXNet for OSX
<dependency>
<groupId>ai.djl.spring</groupId>
<artifactId>djl-spring-boot-starter-mxnet-osx-x86_64</artifactId>
<version>${djl.starter.version}</version>
</dependency>
MXNet for Linux
<dependency>
<groupId>ai.djl.spring</groupId>
<artifactId>djl-spring-boot-starter-mxnet-linux-x86_64</artifactId>
<version>${djl.starter.version}</version>
</dependency>
MXNet Auto Configuration
This will download the correct artifact at runtime (provided external internet egress is enabled), including Windows specific artifacts:
<dependency>
<groupId>ai.djl.spring</groupId>
<artifactId>djl-spring-boot-starter-mxnet-auto</artifactId>
<version>${djl.starter.version}</version>
</dependency>
Note: with automatic configuration, the target runtime environment requires access to the internet because the required platform dependency is downloaded from the maven repository at runtime. This may not be ideal for production environments.
Auto configuration for PyTorch (the correct OS specific artifact is downloaded at runtime):
<dependency>
<groupId>ai.djl.spring</groupId>
<artifactId>djl-spring-boot-starter-pytorch-auto</artifactId>
<version>${djl.starter.version}</version>
</dependency>
Auto configuration for TensorFlow (the correct OS specific artifact is downloaded at runtime):
<dependency>
<groupId>ai.djl.spring</groupId>
<artifactId>djl-spring-boot-starter-tensorflow-auto</artifactId>
<version>${djl.starter.version}</version>
</dependency>
The project also provides Spring autoconfiguration, which completes inference configuration by automatically configuring the model and predictor.
To use autoconfiguration, use the following dependency:
<dependency>
<groupId>ai.djl.spring</groupId>
<artifactId>djl-spring-boot-starter-autoconfigure</artifactId>
<version>${djl.starter.version}</version>
</dependency>
Autoconfiguration supports properties that help with automatic lookup of the model.
For example (assuming application.yml is used):
djl:
# Define application type
application-type: OBJECT_DETECTION
# Define input data type, a model may accept multiple input data type
input-class: java.awt.image.BufferedImage
# Define output data type, a model may generate different out put
output-class: ai.djl.modality.cv.output.DetectedObjects
# Define filters that matches your application's need
model-filter:
size: 512
backbone: mobilenet1.0
# Override default pre-processing/post-processing behavior
arguments:
threshold: 0.2
This configuration has IDE level support for content assistance in Intellij IDEA, Eclipse (with STS), and Netbeans. For more information on available criteria that are currently part of the repository, see the DJL - MXNet model zoo.
See djl-spring-boot-console-sample.
For a more advanced example of the starter's capability, see the DJL Spring Boot Demo.
The demo project contains instructions how to deploy DJL based microservices to Amazon EKS (Elastic Kubernetes Service) and PCF (Pivotal Cloud Foundry).
Also, please see the following post on the general applicability and usage: Adopting machine learning in your microservices with DJL (Deep Java Library) and Spring Boot.
This project is licensed under the Apache-2.0 License.