Ubuntu

Overview

This section describes how to enable MediaTek AI development capabilities on Ubuntu by installing the mtk-neuropilot snap package. With this snap, developers can use the same Neuron tools and workflows that are available on IoT Yocto, and can follow the Yocto documentation for model conversion, deployment, and profiling.

Ubuntu for MediaTek Genio

Canonical provides Ubuntu images optimized for MediaTek Genio platforms. Developers can download supported Ubuntu images and related documentation from the official Ubuntu for MediaTek Genio page.

Ensure that the target board runs a supported Ubuntu release from this page before installing the mtk-neuropilot snap.

Installing NeuroPilot on Ubuntu

The mtk-neuropilot package is distributed via Snap Store. On a supported Ubuntu system with snapd enabled, install it with:

snap install mtk-neuropilot --edge

Note

The --edge channel delivers the latest features and updates, and may change more frequently than stable channels. For production use, refer to the Snap Store page for available channels and their support status.

After installation, the Neuron tools (for example, ncc-tflite and neuronrt) are available inside the snap environment and can be used for model conversion and on‑device inference workflows that are consistent with IoT Yocto.

Using Ubuntu with Genio AI Workflows

When mtk-neuropilot is installed on Ubuntu, developers can:

  • Prepare and convert models (for example, PyTorch → TFLite → DLA) using the same tools and command‑line options described in the IoT Yocto documentation.

  • Use Neuron Compiler (ncc-tflite) to generate Deep Learning Archive (DLA) files.

  • Use Neuron Runtime (neuronrt) and the Neuron Runtime API to test and profile compiled models.

  • Follow the same troubleshooting, performance tuning, and best‑practice flows documented for Genio Yocto.

In other words, the core AI toolchain and workflows are aligned between Ubuntu (via the mtk-neuropilot snap) and IoT Yocto. Developers can reuse the Yocto documentation as a reference when building, converting, and deploying models on Ubuntu.

Referencing Yocto Documentation

For detailed usage of the Neuron tools and Genio AI workflows, refer to the following sections in the IoT Yocto documentation:

The commands, options, and recommended practices described in these sections also apply when working with the mtk-neuropilot snap on Ubuntu, unless explicitly noted otherwise.