Skip to main content
Version: Torizon OS 7.x.y

Machine Learning Documentation Overview

Introduction​

This section of the documentation provides resources and guides for developing Machine Learning applications on Toradex System-on-Modules (SoMs). It also includes information on the principal Machine Learning frameworks and tools supported by Torizon OS, as well as best practices for optimizing Machine Learning models for edge devices.

The most efficient way to execute Machine Learning applications on Toradex SoMs is by leveraging Neural Processing Units (NPUs). NPUs are specialized hardware accelerators designed to efficiently execute the core operations used in neural networks, such as matrix multiplications and convolutions. By offloading these compute-intensive tasks to the NPU, you can significantly reduce inference time, improve energy efficiency, and achieve better overall performance.

NPU Support on Toradex SoMs​

Currently, the NPU is supported on the following Toradex SoMs:

Torizon-Supported Frameworks​

The table below summarizes the current status of Machine Learning frameworks on Torizon OS, and the availability of containers for each framework runtime. Keep in mind that the availability of certain frameworks depend on SoC vendor, may be limited by hardware capabilities, or may be under active development or planned for future releases.

Table Legend:

  • ✅: Framework is supported
  • ❌: Framework is not supported
  • ➖: Framework not available due to hardware limitations or cannot be supported (usually because it requires hardware capabilities not provided by the SoM)
  • ⌛: Framework currently under development or scheduled for an upcoming release
  • 📅: Framework planned for a future release
Platform →
Framework ↓		AM69iMX95iMX8M PlusSL1680
ONNX Runtime📅📅📅📅
OpenCV📅📅📅📅
NNStreamer📅📅📅📅
TensorFlow Lite📅📅✅⌛
LiteRT📅📅📅📅
Torq Runtime*➖➖➖📅

*Torq Runtime is a proprietary software runtime developed by Synaptics. It is not supported on other platforms due to hardware limitations.

If you are interested in using a framework that is not supported on Torizon OS or not listed in the table above, contact us.

Use Cases​

Legacy Documentation​

The Developer Website provides Machine Learning guides for older versions of Torizon OS. Although these guides may still serve as useful examples, they have not been updated to reflect the latest changes in the software stack. These articles include:

Send Feedback!