2. Installation

Tiny ML ModelZoo 1.5.0 installs with pip. There are exactly two supported install paths, and neither needs more than one command to get the whole toolchain (tinyml_modelmaker, tinyml_tinyverse and tinyml_torchmodelopt come in automatically as prebuilt wheels from TI’s download server).

2.9. Which Install Should I Choose?

Scenario

Use This Install

I want to train and compile with the TI-provided models, using my own config files

Wheel install (Wheel Installation)

I want to use the model zoo as a Python library in my own scripts

Wheel install (Wheel Installation)

I want the ready-to-run examples/ folder and the run_tinyml_modelzoo scripts

Clone install (Clone Installation)

I want to add or change a model architecture

Clone install (Clone Installation)

I want to bring my own model (BYOM)

Clone install (Clone Installation)

2.10. Quick Start

pip install https://software-dl.ti.com/C2000/esd/mcu_ai/wheel/tinyml_modelzoo-1.5.0-py3-none-any.whl

Note

Starting with 1.5.0 you clone only tinyml-modelzoo. You no longer clone tinyml-tensorlab or the separate modelmaker, tinyverse and modeloptimization repositories, and there is no setup_all script. tinyml-tensorlab is the 1.4.0 entry point and is now deprecated.

After installing, continue with Environment Variables (compiler setup, needed only for compiling to a device) and then Quickstart.