2.1. Prerequisites

Check these requirements before installing Tiny ML ModelZoo 1.5.0.

2.1.1. System Requirements

Linux

Ubuntu 20.04 or later recommended (x86_64)

Windows

Windows 10/11, native or WSL2

macOS

Apple Silicon (arm64) with MPS training support. See macOS Setup.

CPU / RAM / disk

Multi-core CPU, 8 GB RAM minimum (16 GB recommended), about 10 GB free disk for the install and datasets

GPU

Optional. CUDA GPU on Linux/Windows, or MPS on Apple Silicon, to speed up training

2.1.2. Python

Important

Use Python 3.10 if you want to compile models for a device.

  • tinyml_modelzoo declares requires-python >=3.10,<3.15 and 1.5.0 added Python 3.14 / PyTorch 2.11 compatibility, so training works on Python 3.10 to 3.14.

  • The TI Neural Network Compiler wheel (ti_mcu_nnc 2.1.2) that ModelMaker uses for compilation is only installed on Python 3.10 (its dependency marker is python_version < '3.11'). On Python 3.11 or later, pip skips it silently and the compile step is not available.

  • The project README lists “Python 3.10 environment” as the prerequisite.

Pinned dependencies installed by pip (from pyproject.toml): torch==2.11.0, PyYAML==6.0.3, numpy==2.2.6 (Python 3.10) or numpy==2.5.2 (Python 3.11+). Do not pre-install other versions of torch or numpy in the same environment.

pyenv is a convenient way to get Python 3.10:

curl https://pyenv.run | bash       # macOS: brew install pyenv
pyenv install 3.10.14
pyenv global 3.10.14
python --version

2.1.3. pip and Git

python -m pip install --upgrade pip setuptools wheel
git --version        # only needed for the clone install

Git is only needed for the clone install (Clone Installation). The wheel install (Wheel Installation) needs only Python and pip.

2.1.4. TI Compiler Tools (for compilation)

Required only to compile models for a device. Download from TI and set the matching environment variable (Environment Variables).

To flash and debug on hardware, install Code Composer Studio (https://www.ti.com/tool/CCSTUDIO) and see Deploying a ModelZoo Run to an SDK Example (Golden-Vector Check).

2.1.5. Verification Checklist

python --version     # 3.10.x for compilation support
pip --version
git --version        # clone install only

Then proceed to Wheel Installation or Clone Installation.