Installation

Important

MetaTF 2.16 was the last release supporting TensorFlow 2.15.0 (Keras 2) and Python 3.9.

Starting with MetaTF 2.17, releases will support TF-Keras 2.19 (TensorFlow 2.19) and Python 3.10 to 3.12.

Supported configurations

  • Operating systems:

    • Windows 10, Windows 11

    • Any Linux variant compatible with manylinux 2.28 (Ubuntu 22.04, Ubuntu 24.04, …)

  • Python versions: 3.10 to 3.12

  • TF-Keras versions: 2.19

  • PyTorch versions: While not an explicit MetaTF/ONNX requirement, PyTorch versions >= 2.6 are supported. GPU support and TensorFlow dependency management are left to the user’s discretion.

Warning

On Windows, the latest Visual C++ redistributable package is required. Please refer to this link for installation.

Quick installation

The complete MetaTF framework can be set up with Python’s pip package manager by installing the metatf package only:

pip install metatf==2.19.3

Note

metatf is a meta-package that pulls the complete, validated set of MetaTF packages as dependencies: akida-models, cnn2snn, akida and quantizeml, along with TensorFlow and TF-Keras — there is no need to install TensorFlow separately. metatf itself contains no code: it provides no metatf module or command line, the four packages above are imported and used directly. For a GPU-enabled TensorFlow setup, please refer to Install TensorFlow with pip.

Note

We recommend using a virtual environment such as Conda. Please note that the Python version must be explicitly specified when creating a conda environment. The specification must be for one of the supported Python versions listed above.

conda create --name akida_env python=3.11
conda activate akida_env

Verify the installation

Once the packages are installed, check that everything is in place:

python -c "import akida; print(akida.__version__); print(akida.devices())"

The expected output is:

2.19.3
[]

An empty device list [] is the correct result on a machine without Akida hardware: no hardware is required, and everything on this site (user guide and examples) runs on the software simulator that comes with the akida package.

Note

If the import fails with ModuleNotFoundError, the virtual environment where the packages were installed is probably not active. If the printed version is not 2.19.3, update the framework with pip install --upgrade metatf==2.19.3.

Running examples

The Akida tutorials can be downloaded from the examples section as Python scripts or Jupyter Notebooks. Dependencies needed to replay the examples can be installed using the requirements.txt file:

pip install -r requirements.txt

Note

Please refer to this link for Jupyter Notebook installation and configuration.