DJL - Python engine implementation

Overview

This module contains the Deep Java Library (DJL) EngineProvider for Python based model.

DJL Python engine allows you run python model in a JVM based application. However, you still need to install your python environment and dependencies.

Python engine is a DL library with limited support for NDArray operations. Currently, it only covers the basic NDArray creation methods. To better support the necessary preprocessing and postprocessing, you can use one of the other Engine along with it to run in a hybrid mode. For more information, see Hybrid Engine.

Documentation

The latest javadocs can be found on javadoc.io.

You can also build the latest javadocs locally using the following command:

# for Linux/macOS:
./gradlew javadoc

# for Windows:
..\..\gradlew javadoc

The javadocs output is generated in the build/doc/javadoc folder.

Installation

You can pull the Python engine from the central Maven repository by including the following dependency:

• ai.djl.python:python:0.23.0

<dependency>
    <groupId>ai.djl.python</groupId>
    <artifactId>python</artifactId>
    <version>0.23.0</version>
    <scope>runtime</scope>
</dependency>

Test your python model

Testing python code within Java environment is challenging. We provide a tool to help you develop and test your python model locally. You can easily use IDE to debug your model.

1. Install djl_python toolkit:

cd engines/python/setup
pip install -U -e .

2. You can use command line tool or python to run djl model testing. The following command is an example:

curl -O https://resources.djl.ai/images/kitten.jpg

djl-test-model --model-dir src/test/resources/resnet18 --input kitten.jpg

# or use python
python -m djl_python.test_model --model-dir src/test/resources/resnet18 --input kitten.jpg