Install TensorFlow Java

TensorFlow Java can run on any JVM for building, training and deploying machine learning models. It supports both CPU and GPU execution, in graph or eager mode, and presents a rich API for using TensorFlow in a JVM environment. Java and other JVM languages, like Scala and Kotlin, are frequently used in large and small enterprises all over the world, which makes TensorFlow Java a strategic choice for adopting machine learning at a large scale.


TensorFlow Java runs on Java 8 and above, and supports out-of-the-box the following platforms:

  • Ubuntu 16.04 or higher; 64-bit, x86
  • macOS 10.12.6 (Sierra) or higher; 64-bit, x86
  • Windows 7 or higher; 64-bit, x86


TensorFlow Java has its own release cycle, independent from the TensorFlow runtime. Consequently, its version does not match the version of TensorFlow runtime it runs on. Consult the TensorFlow Java versioning table to list all versions available and their mapping with the TensorFlow runtime.


There are several ways to add TensorFlow Java to your project. The easiest one is to add a dependency on the tensorflow-core-platform artifact, which includes both the TensorFlow Java Core API and the native dependencies it requires to run on all supported platforms.

You can also select one of the following extensions instead of the pure CPU version:

  • tensorflow-core-platform-mkl: Support for Intel® MKL-DNN on all platforms
  • tensorflow-core-platform-gpu: Support for CUDA® on Linux and Windows platforms
  • tensorflow-core-platform-mkl-gpu: Support for Intel® MKL-DNN and CUDA® on Linux platform.

In addition, a separate dependency on the tensorflow-framework library can be added to benefit from a rich set of utilities for TensorFlow-based machine learning on the JVM.

Installing with Maven

To include TensorFlow in your Maven application, add a dependency on its artifacts to your project's pom.xml file. For example,


Reducing Number of Dependencies

It is important to note that adding a dependency on a tensorflow-core-platform artifact will import native libraries for all supported platforms, which can significantly increase the size of your project.

If you wish to target a subset of the available platforms then you can exclude the unnecessary artifacts from the other platforms using the Maven Dependency Exclusion feature.

Another way to select which platforms you want to include in your application is to set JavaCPP system properties, in your Maven command line or in your pom.xml. Please see JavaCPP documentation for more details.

Using Snapshots

The latest TensorFlow Java development snapshots from the TensorFlow Java source repository are available on the OSS Sonatype Nexus repository. To depend on these artifacts, make sure to configure the OSS snapshots repository in your pom.xml.



Installing with Gradle

To include TensorFlow in your Gradle application, add a dependency on its artifacts to your project's build.gradle file. For example,

repositories {

dependencies {
    compile group: 'org.tensorflow', name: 'tensorflow-core-platform', version: '0.3.3'

Reducing Number of Dependencies

Excluding native artifacts from TensorFlow Java with Gradle is not as easy as with Maven. We recommend that you use Gradle JavaCPP plugins to reduce this number of dependencies.

Please read at Gradle JavaCPP documentation for more details.

Installing from Sources

To build TensorFlow Java from sources, and possibly customize it, please read the following instructions.

Example Program

This example shows how to build an Apache Maven project with TensorFlow. First, add the TensorFlow dependency to the project's pom.xml file:


        <!-- Minimal version for compiling TensorFlow Java is JDK 8 -->

        <!-- Include TensorFlow (pure CPU only) for all supported platforms -->

Create the source file src/main/java/

import org.tensorflow.ConcreteFunction;
import org.tensorflow.Signature;
import org.tensorflow.Tensor;
import org.tensorflow.TensorFlow;
import org.tensorflow.op.Ops;
import org.tensorflow.op.core.Placeholder;
import org.tensorflow.op.math.Add;
import org.tensorflow.types.TInt32;

public class HelloTensorFlow {

  public static void main(String[] args) throws Exception {
    System.out.println("Hello TensorFlow " + TensorFlow.version());

    try (ConcreteFunction dbl = ConcreteFunction.create(HelloTensorFlow::dbl);
        TInt32 x = TInt32.scalarOf(10);
        Tensor dblX = {
      System.out.println(x.getInt() + " doubled is " + ((TInt32)dblX).getInt());

  private static Signature dbl(Ops tf) {
    Placeholder<TInt32> x = tf.placeholder(TInt32.class);
    Add<TInt32> dblX = tf.math.add(x, x);
    return Signature.builder().input("x", x).output("dbl", dblX).build();

Compile and execute:

mvn -q compile exec:java

The command prints TensorFlow version and a simple calculation.

Success! TensorFlow Java is configured.