model-inference-table1-2

This example evaluates the inference speed on microcontrollers natively, and with WAMR interpreter and WAMR AoT (Ahead-of-Time compilation). The example uses RIOT-ML to generate the machine learning inference code.

Setup RIOT_ML

You must install Apache TVM to be used by RIOT-ML. To do this, you can use an automated script available on:

bash ../../../scripts/install-RIOT-ML.bash

Generating native model example

To generate native model example, run in the current folder:

../../../scripts/create_example_tinyml.bash
cp -r model-template/* model

Usage

In each subfolder, you can:

• run BOARD=arduino-nano-33-ble make clean all flash term to flash the inference and evaluation code on your board. It will display the time required to do inference with DS CNN model.
  • You can replace arduino-nano-33-ble with your board, the full list of supported boards is available in RIOT-OS documentation.
• run BOARD=arduino-nano-33-ble make clean bench-size to display the size of the different components in the produced image.