Machine Learning Library Extensions

Overview

These extensions bundle pip packages. Using these extensions as a dependency saves you save development time setting up the dependency, reduces your extension’s size during the build process, and reduces clutter on users’ devices. Versions of the ML libraries are defined by tags, so switching versions only requires a tag change. Frameworks we currently support:

Installation

The machine learning libraries are intended to be used by developers authoring kit extensions.

We plan to regularly release support for new versions of these libraries. To use one of the libraries simply add a tagged version of one the following kit extensions as a dependency (LINK to explain dependency):

  • PyTorch (omni.pip.torch)
    • Current LTS (1.8.2): "omni.pip.torch" = {tag="lts"}

    • 1.11.0: "omni.pip.torch" = {tag="1_11_0"}

    • 1.12.0: "omni.pip.torch" = {tag="1_12_0"}

  • ONNX (omni.pip.onxx)
    • 1.11.1: "omni.pip.onnx" = {tag="1_11_1"}

  • TensorRT (omni.pip.tensorrt)
    • The TensorRT extension includes the matching runtime dependencies for inference.

    • 8.2.5: "omni.pip.tensorrt" = {tag="8_2_5"}

    • 8.4.2: "omni.pip.tensorrt" = {tag="8_4_2"}. Due to current releases this provides version 8.4.2.4 on Windows and 8.4.1.5 on Linux.

Using the Machine Learning Libraries

With these extensions installed the corresponding python libraries may be used as in any python script. Simply import from the library and call the functions as desired.