TensorFlow can run on any JVM for building, training and running machine learning models. It comes with a series of utilities and frameworks that help achieve most of the tasks common to data scientists and developers working in this domain. Java and other JVM languages, such as Scala or Kotlin, are frequently used in small-to-large enterprises all over the world, which makes TensorFlow a strategic choice for adopting machine learning at a large scale.

In the early days, the Java language bindings for TensorFlow were hosted in the main repository and released only when a new version of the core library was ready to be distributed, which happens only a few times a year. Now, all Java-related code has been moved to this repository so that it can evolve and be released independently from official TensorFlow releases. In addition, most of the build tasks have been migrated from Bazel to Maven, which is more familiar for most Java developers.

The following describes the layout of the repository and its different artifacts:

• tensorflow-core

  • All artifacts that build up the core language bindings of TensorFlow for Java
  • Intended audience: projects that provide their own APIs or frameworks on top of TensorFlow and just want a thin layer to access the TensorFlow native library from the JVM

• tensorflow-framework

  • Primary API for building and training neural networks with TensorFlow
  • Intended audience: neural network developers
  • For more information: tensorflow-framework/README.md

Note: The NdArray Library module has now its own repository and has been moved out of TensorFlow Java.

This repository is maintained by TensorFlow JVM Special Interest Group (SIG). You can easily contact the group by posting to the TensorFlow Forum, adding the sig_jvm tag, or by writing to us on the sig-jvm Gitter channel. You can also simply send pull requests and raise issues to this repository.

See CONTRIBUTING.md.

There are two options for adding TensorFlow Java as a dependency to your Maven project: with individual dependencies for each targeted platforms or with a single dependency that target them all.

With this option, you must first add a dependency to tensorflow-core-api and then one or multiple dependencies to tensorflow-core-native with a classifier targeting a specific platform. This option is preferred as it minimize the size of your application by only including the TensorFlow builds you need, at the cost of being more restrictive.

While TensorFlow Java can be compiled for multiple platforms, only binaries for the followings are being supported and distributed by this project:

• linux-x86_64: Linux platforms on Intel chips
• linux-x86_64-gpu: Linux platforms on Intel chips with Cuda GPU support
• macosx-x86_64: MacOS X platforms on Intel chips
• macosx-arm64: MacOS X platforms on Apple Silicon chips
• windows-x86_64: Windows platforms on Intel chips

For example, for building a JAR that uses TensorFlow and is targeted to be deployed only on Linux systems with no GPU support, you should add the following dependencies:

<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-api</artifactId>
  <version>0.5.0</version>
</dependency>
<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-native</artifactId>
  <version>0.5.0</version>
  <classifier>linux-x86_64</classifier>
</dependency>

On the other hand, if you plan to deploy your JAR on more platforms, you need additional native dependencies as follows:

<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-api</artifactId>
  <version>0.5.0</version>
</dependency>
<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-native</artifactId>
  <version>0.5.0</version>
  <classifier>linux-x86_64-gpu</classifier>
</dependency>
<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-native</artifactId>
  <version>0.5.0</version>
  <classifier>macosx-arm64</classifier>
</dependency>
<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-native</artifactId>
  <version>0.5.0</version>
  <classifier>windows-x86_64</classifier>
</dependency>

Only one dependency can be added per platform, meaning that you cannot add native dependencies to both linux-x86_64 and linux-x86_64-gpu within the same project.

In some cases, it might be preferable to add a single dependency that includes transitively all the artifacts required to run TensorFlow Java on any supported platforms

• tensorflow-core-platform: Includes tensorflow-core-api, plus native artifacts for linux-x86_64, macosx-arm64, macosx-x86_64 and windows-x86_64

For example, to run TensorFlow Java on any platform for which a binary is being distributed by this project, you can simply add this dependency to your application:

<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-platform</artifactId>
  <version>0.5.0</version>
</dependency>

Be aware though that the builds of TensorFlow are quite voluminous and including too many native dependencies may significantly increase the size of your application. So it is good practice to limit your dependencies to the platforms you are targeting. For this purpose these artifacts include profiles that follow the conventions established on this page:

• Reducing the Number of Dependencies

Snapshots of TensorFlow Java artifacts are automatically distributed after each update in the code. To use them, you need to add Sonatype OSS repository in your pom.xml, like the following

<repositories>
    <repository>
        <id>tensorflow-snapshots</id>
        <url>https://oss.sonatype.org/content/repositories/snapshots/</url>
        <snapshots>
            <enabled>true</enabled>
        </snapshots>
    </repository>
</repositories>
<dependencies>
    <!-- Example of dependency, see section above for more options -->
    <dependency>
        <groupId>org.tensorflow</groupId>
        <artifactId>tensorflow-core-platform</artifactId>
        <version>1.0.0-SNAPSHOT</version>
    </dependency>
</dependencies>

This table shows the mapping between TensorFlow, TensorFlow Java and minimum supported Java versions.

Contributions are welcome, guidelines are located in CONTRIBUTING.md.

Please look at this repository: https://github.com/tensorflow/java-models