This repository is a demonstration of how to use PyTorch from Java.
Download and unpack libtorch nightly (or 1.4 or greater). From the pytorch.org homepage under "Quick Start Locally", make sure "LibTorch" is the selected package. Linux is supported as of version 1.4. Mac is supported as of version 1.5. Windows is supported as of version 1.9. All are both supported in nightly builds.
Run export LIBTORCH_HOME=/path/to/libtorch.
The build.gradle file will use this to set java.library.path
when running the application.
If you are using PyTorch in your own environment,
LIBTORCH_HOME is not necessary.
Instead, you will need to set java.library.path to /path/to/libtorch/lib.
It might be necessary to set the JAVA_HOME environment variable.
On Mac, use export JAVA_HOME=$(/usr/libexec/java_home).
On Windows, with GitHub bash, try export JAVA_HOME=/c/Progra~1/Java/jdk-*
On Mac, you might get security errors.
You must manually approve execution for each library under libtorch.
As a shortcut, xattr -r -d /path/to/libtorch/lib might work.
On Windows, you will probably need to run
export PATH="$LIBTORCH_HOME/lib:$PATH".
If using a nightly build, run export USE_LIBTORCH_NIGHTLY=1.
Run ./gradlew run to build and run the demo application.
It will load demo-model.pt and run it on some simple data.
This notebook was used to generate the model.
More information about the Java API and TorchScript: