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 runtime 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
-
ndarray- Generic utility library for n-dimensional data I/O operations
- Used by TensorFlow but does not depend on TensorFlow
- Intended audience: any developer who needs a Java n-dimensional array implementation, whether or not they use it with TensorFlow
To build all the artifacts, simply invoke the command mvn install at the root of this repository (or
the Maven command of your choice). It is also possible to build artifacts with support for MKL enabled with
mvn install -Djavacpp.platform.extension=-mkl or CUDA with mvn install -Djavacpp.platform.extension=-gpu
or both with mvn install -Djavacpp.platform.extension=-mkl-gpu.
When building this project for the first time in a given workspace, the script will attempt to download the TensorFlow runtime library sources and build of all the native code for your platform. This requires a valid environment for building TensorFlow, including the bazel build tool and a few Python dependencies (please read TensorFlow documentation for more details).
This step can take multiple hours on a regular laptop. It is possible though to skip completely the native build if you are
working on a version that already has pre-compiled native artifacts for your platform available on Sonatype OSS Nexus repository.
You just need to activate the dev profile in your Maven command to use those artifacts instead of building them from scratch
(e.g. mvn install -Pdev).
Note that modifying any source files under tensorflow-core may impact the low-level TensorFlow bindings, in which case a
complete build could be required to reflect the changes.
To include TensorFlow in your Maven application, you first need to add a dependency on either the
tensorflow-core or tensorflow-core-platform artifacts. The former could be included multiple times
for different targeted systems by their classifiers, while the later includes them as dependencies for
linux-x86_64, macosx-x86_64, and windows-x86_64, with more to come in the future. There are also
tensorflow-core-platform-mkl, tensorflow-core-platform-gpu, and tensorflow-core-platform-mkl-gpu
artifacts that depend on artifacts with MKL and/or CUDA support enabled.
For example, for building a JAR that uses TensorFlow and is targeted to be deployed only on Linux systems, you should add the following dependencies:
<dependency>
<groupId>org.tensorflow</groupId>
<artifactId>tensorflow-core-api</artifactId>
<version>0.2.0</version>
</dependency>
<dependency>
<groupId>org.tensorflow</groupId>
<artifactId>tensorflow-core-api</artifactId>
<version>0.2.0</version>
<classifier>linux-x86_64${javacpp.platform.extension}</classifier>
</dependency>On the other hand, if you plan to deploy your JAR on more platforms, you need additional native dependencies as follows:
<dependency>
<groupId>org.tensorflow</groupId>
<artifactId>tensorflow-core-api</artifactId>
<version>0.2.0</version>
</dependency>
<dependency>
<groupId>org.tensorflow</groupId>
<artifactId>tensorflow-core-api</artifactId>
<version>0.2.0</version>
<classifier>linux-x86_64${javacpp.platform.extension}</classifier>
</dependency>
<dependency>
<groupId>org.tensorflow</groupId>
<artifactId>tensorflow-core-api</artifactId>
<version>0.2.0</version>
<classifier>macosx-x86_64${javacpp.platform.extension}</classifier>
</dependency>
<dependency>
<groupId>org.tensorflow</groupId>
<artifactId>tensorflow-core-api</artifactId>
<version>0.2.0</version>
<classifier>windows-x86_64${javacpp.platform.extension}</classifier>
</dependency>In some cases, pre-configured starter artifacts can help to automatically include all versions of
the native library for a given configuration. For example, the tensorflow-core-platform,
tensorflow-core-platform-mkl, tensorflow-core-platform-gpu, or tensorflow-core-platform-mkl-gpu
artifact includes transitively all the artifacts above as a single dependency:
<dependency>
<groupId>org.tensorflow</groupId>
<artifactId>tensorflow-core-platform${javacpp.platform.extension}</artifactId>
<version>0.2.0</version>
</dependency>Be aware though that the native library is quite large and including too many versions of it may
significantly increase the size of your JAR. So it is good practice to limit your dependencies to
the platforms you are targeting. For this purpose the -platform artifacts include profiles that follow
the conventions established on this page:
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>0.3.0-SNAPSHOT</version>
</dependency>
</dependencies>This table shows the mapping between different version of TensorFlow for Java and the core runtime libraries.
| TensorFlow Java Version | TensorFlow Version |
|---|---|
| 0.2.0 | 2.3.1 |
| 0.3.0-SNAPSHOT | 2.3.1 |
This repository is maintained by TensorFlow JVM Special Interest Group (SIG). You can easily join the group by subscribing to the [email protected] mailing list, or you can simply send pull requests and raise issues to this repository.