QuantLib

QuantLib is a library to train deploy quantised neural networks (QNNs). It was developed on top of the PyTorch deep learning framework.

QuantLib is a component of QuantLab, which also includes organising software to manage machine learning (ML) experiments (systems and manager packages, as well as the main.py façade script).

Installation and usage

Create an Anaconda environment and install quantlib

Use Anaconda or Miniconda to install QuantLab's prerequisites.

PyTorch 1.13.1 (Recommended)

$> conda create --name pytorch-1.13
$> conda activate pytorch-1.13
$> conda config --env --add channels conda-forge
$> conda config --env --add channels pytorch 
$> conda install python=3.8 pytorch=1.13.1 pytorch-gpu torchvision=0.14.1 torchtext=0.14.1 torchaudio-0.13.1 cudatoolkit=11.6 -c pytorch -c conda-forge
$> conda install ipython packaging parse setuptools tensorboard tqdm networkx python-graphviz scipy pandas ipdb onnx onnxruntime einops yapf tabulate
$> pip install setuptools==59.5.0 torchsummary parse coloredlogs netron

PyTorch 1.12.1

$> conda create --name pytorch-1.12
$> conda activate pytorch-1.12
$> conda config --env --add channels conda-forge
$> conda config --env --add channels pytorch 
$> conda install python=3.8 pytorch=1.12.1 torchvision=0.13.1 torchtext=0.13.1 torchaudio=0.12.1 -c pytorch -c conda-forge
$> conda install ipython packaging parse setuptools tensorboard tqdm networkx python-graphviz scipy pandas ipdb onnx onnxruntime einops yapf tabulate
$> pip install setuptools==59.5.0 torchsummary parse coloredlogs netron

After creating the conda environment, install the quantlib quantisation library in your Anaconda environment:

$ conda activate quantlab
(quantlab) $ cd quantlib
(quantlab) $ python setup.py install
(quantlab) $ cd ..

Notice

Licensing information

quantlib is distributed under the Apache 2.0 license.

In case you are planning to use QuantLab and quantlib in your projects, you might also want to consider the licenses under which the packages on which they depend are distributed:

• PyTorch - a mix of licenses, including the Apache 2.0 License and the 3-Clause BSD License;
• TensorBoard - Apache 2.0 License;
• NetworkX - 3-Clause BSD License;
• GraphViz - MIT License;
• matplotlib - a custom license;
• NumPy - 3-Clause BSD License;
• SciPy - 3-Clause BSD License;
• Mako - MIT License;
• Jupyter - 3-Clause BSD License.

Authors

• Matteo Spallanzani <[email protected]> (ETH Zurich, now at Axelera AI)
• Georg Rutishauser <[email protected]> (ETH Zurich)
• Moritz Scherer <[email protected]> (ETH Zurich)
• Philip Wiese <[email protected]> (ETH Zurich)
• Francesco Conti <[email protected]> (University of Bologna)