2.2. Wheel Installation

Use this path if you plan to make no model additions or changes. It is one pip install of the tinyml_modelzoo wheel.

2.2.1. Quick Install

Create and activate a virtual environment (see Prerequisites), then run:

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

The wheel declares these dependencies, so pip installs them for you:

  • tinyml_modelmaker 1.5.0 (downloaded from the same TI server). It in turn pulls in tinyml_tinyverse (training), tinyml_torchmodelopt (quantization) and ti_mcu_nnc (the TI Neural Network Compiler).

  • torch==2.11.0, PyYAML==6.0.3 and a pinned numpy.

2.2.2. What Is in the Wheel

The wheel contains only the Python package tinyml_modelzoo: the model definitions (tinyml_modelzoo.models), the model descriptions that map the training.model_name values in a config to models, and the custom layers. It does not contain:

  • the examples/ folder with ready-to-run configs, and

  • the run_tinyml_modelzoo.sh / .bat / .ps1 wrapper scripts.

This is by design (it is a Python package), but it means a wheel-only user supplies their own config file.

2.2.3. Running with the Wheel Install

Write a config YAML (see How to Create a ModelZoo Config YAML), or download one of the example configs from the examples folder on GitHub (for instance examples/generic_timeseries_classification/config.yaml). Then call the ModelMaker entry point directly. This is exactly what the wrapper scripts do:

python -m tinyml_modelmaker.run_tinyml_modelmaker config.yaml

Output is written to ./data/projects/<project_name>/run/<date-time>/<model_name>/ relative to the folder you run the command from; set common.projects_path in the config to change it. Datasets named in the config are downloaded automatically.

Warning

To compile for a device, set the compiler environment variable for your target before running. See Environment Variables. Training and quantization work without it.

2.2.4. Verifying Installation

import tinyml_modelmaker
import tinyml_tinyverse
import tinyml_torchmodelopt
import tinyml_modelzoo

print(f"TI Tiny ML ModelMaker: {tinyml_modelmaker.__version__}")
print(f"TI Tiny ML Tinyverse: {tinyml_tinyverse.__version__}")
print(f"TI Tiny ML Model Optimization toolkit: {tinyml_torchmodelopt.__version__}")
print(f"TI Tiny ML Model Zoo: {tinyml_modelzoo.__version__}")

All four should print 1.5.0-series version numbers.

2.2.5. Updating and Uninstalling

To move to a newer release, install the newer wheel URL the same way (add --upgrade). To remove the toolchain:

pip uninstall tinyml-modelzoo tinyml-modelmaker tinyml-tinyverse tinyml-torchmodelopt

2.2.6. Limitations of the Wheel Install

  • You cannot add or edit models without reinstalling (use Clone Installation).

  • The shipped examples/ and wrapper scripts are not included.

2.2.7. Troubleshooting

“No module named tinyml_modelmaker”

You are in a different Python environment than the one you installed into. Check which python (Linux/macOS) or where python (Windows) and python --version.

ti_mcu_nnc is not installed / compilation is unavailable

The TI Neural Network Compiler wheel is published for Python 3.10 only. On Python 3.11 or later training works, but compilation does not. See Prerequisites.

Version conflicts

Install into a fresh virtual environment:

python -m venv venv
source venv/bin/activate      # Windows: venv\Scripts\activate
pip install https://software-dl.ti.com/C2000/esd/mcu_ai/wheel/tinyml_modelzoo-1.5.0-py3-none-any.whl

2.2.8. Next Steps