Installation

Supported configurations

  • Operating systems:

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

  • Python versions: 3.10 to 3.12

  • cnn2snn version: 2.19.0

  • onnxruntime version: 1.19.2

  • pytorch version: While not an explicit MetaTF/ONNX requirement, Pytorch versions >= 2.6 are supported. GPU support and Tensorflow dependency management are left to the user discretion.

Quick installation

MetaONNX relies on onnx2akida python package that can be setup with python’s pip package manager:

pip install onnx2akida==0.7.0

MetaONNX is, just as MetaTF, built on top of Brainchip’s usual set of tools that is QuantizeML, CNN2SNN and Akida.

Note

We recommend using 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.

Using Conda:

conda create --name onnx2akida_venv python=3.11
conda activate onnx2akida_venv

Using python venv:

python3.11 -m venv onnx2akida_venv
source onnx2akida_venv/bin/activate

Running examples

The onnx2akida 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.