Class TFJ_RuntimeLibrary
The Java TensorFlow bindings require a native (JNI) library. This library (libtensorflow_jni.so on Linux, libtensorflow_jni.dylib on OS X, tensorflow_jni.dll on Windows) can be made available to the JVM using the java.library.path System property (e.g., using -Djava.library.path command-line argument). However, doing so requires an additional step of configuration.
Alternatively, the native libraries can be packaged in a .jar, making them easily usable from build systems like Maven. However, in such cases, the native library has to be extracted from the .jar archive.
NativeLibrary.load() takes care of this. First looking for the library in java.library.path and failing that, it tries to find the OS and architecture specific version of the library in the set of ClassLoader resources (under org/tensorflow/native/OS-ARCH). The resources paths used for lookup must be consistent with any packaging (such as on Maven Central) of the TensorFlow Java native libraries.
-
Method Summary
-
Method Details
-
load
public static void load()
-