2.6. macOS Setup

macOS-specific notes for Tiny ML ModelZoo 1.5.0.

Note

New in 1.5.0: Apple Silicon (MPS). Release 1.5.0 added macOS arm64 compatibility fixes across training, evaluation and quantization, and training can run on the Apple GPU through PyTorch’s MPS backend.

2.6.1. Supported Hardware

Apple Silicon (arm64) is the supported Mac platform: the compiler wheel (ti_mcu_nnc) used by ModelMaker is published for macOS arm64 (macOS 14 or later, Python 3.10).

2.6.2. Prerequisites

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zprofile
eval "$(/opt/homebrew/bin/brew shellenv)"
brew install git pyenv

Add pyenv to ~/.zshrc:

export PYENV_ROOT="$HOME/.pyenv"
export PATH="$PYENV_ROOT/bin:$PATH"
eval "$(pyenv init -)"

then:

source ~/.zshrc
pyenv install 3.10.14
pyenv global 3.10.14

2.6.3. Installation

python -m venv venv
source venv/bin/activate
python -m pip install --upgrade pip setuptools wheel

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

# or clone install
git clone https://github.com/TexasInstruments/tinyml-modelzoo.git
cd tinyml-modelzoo
pip install -e .

Run the hello world example from the clone:

./run_tinyml_modelzoo.sh examples/generic_timeseries_classification/config.yaml

Set the compiler variable first if you compile (Environment Variables).

2.6.4. Using the Apple GPU (MPS)

Training uses the CPU by default (training.num_gpus: 0). Set num_gpus to 1 or more in the config’s training section: if PyTorch reports MPS as available, training runs on mps; otherwise it uses cuda.

training:
  num_gpus: 1

MPS does not implement every PyTorch operator. ModelMaker sets PYTORCH_ENABLE_MPS_FALLBACK=1 automatically (unless you already set it), so operators that are missing on MPS fall back to the CPU. You do not need to export it yourself.

CUDA is not available on macOS.

2.6.5. Notes

  • Some operations use OpenMP; if you see a libomp error, run brew install libomp.

  • SSL: CERTIFICATE_VERIFY_FAILED while downloading datasets: run the “Install Certificates.command” shipped with the python.org installer, or pip install certifi.

  • pyenv: command not found: reload your shell (source ~/.zshrc).

  • brew: command not found on Apple Silicon: run eval "$(/opt/homebrew/bin/brew shellenv)".

2.6.6. Next Steps