Qiskit-Torch-Module

This repo contains the code for the qiskit-torch-module introduced in "Qiskit-Torch-Module: Fast Prototyping of Quantum Neural Networks", N. Meyer et al., arXiv:2404.06314 (2024).

Quantum computer simulation software is an integral tool for the research efforts in the quantum computing community. An important aspect is the efficiency of respective frameworks, especially for training variational quantum algorithms. Focusing on the widely used Qiskit software environment, we develop the qiskit-torch-module. It improves runtime performance by two orders of magnitude over comparable libraries, while facilitating low-overhead integration with existing codebases. Moreover, the framework provides advanced tools for integrating quantum neural networks with PyTorch. The pipeline is tailored for single-machine compute systems, which constitute a widely employed setup in day-to-day research efforts.

Setup and Installation

The library requires an installation of python 3.12, and following libraries:

• qiskit~=1.0.0, backward compatible up to qiskit v0.44.0
• qiskit-algorithms~=0.3.0
• torch~=2.2.1
• threadpoolctl~=3.3.0

We recommend setting up a conda environment:

conda create --name ENV_NAME python=3.12
conda activate ENV_NAME

The package qiskit-torch-module can be installed locally via:

cd qiskit-torch-module
pip install -e .

Usage and Reproduction of Results

Information on how to use the different modalities of the libraries are described in the documentation. Additionally, we provide two usage examples.

To run the benchmarks and end-to-end implementations, the additional libraries in examples/requirements.txt need to be installed:

• gymnasium~=0.29.1
• qiskit-machine-learning~=0.7.1
• tqdm~=4.66.2
• torchvision~=0.17.1

This can also be done with

pip install -r examples/requirements.txt

Benchmarking of Raw Runtimes

This allows for benchmarking of runtimes for forward and backward pass compared to qiskit-machine-learning:

python examples/benchmark.py --qubits 12 --depth 3 --thread 0 [--use_qml]

This benchmarks the performance for 12 qubits with a circuit depth of 3. Appending / removing the --use_qml flag defines the usage of qiskit-machine-learning / qiskit-torch-module. It is possible to select an explicit number of parallel workers, with threads=0 using all available CPU cores.

End-to-End Quantum Machine Learning Pipelines

Classification

This implements the full-quantum classification algorithm proposed in "Incremental Data-Uploading for Full-Quantum Classification", M. Periyasamy et al., IEEE QCE 1:31-37 (2022)

python examples/qml.py [--use_qml]

Quantum Reinforcement Learning

This implements the quantum policy gradient algorithm proposed in "Quantum Policy Gradient Algorithm with Optimized Action Decoding", N. Meyer et al., PMLR 202:24592-24613 (2023):

python examples/qrl.py [--use_qml]

Acknowledgements

The backbone of our implementation is the qiskit software framework: https://github.com/Qiskit

Furthermore, we got inspired by qiskit-machine-learning: https://github.com/qiskit-community/qiskit-machine-learning

Citation

If you use the qiskit-torch-module or results from the paper, please cite our work as

@article{meyer2024qiskit,
  title={Qiskit-Torch-Module: Fast Prototyping of Quantum Neural Networks},
  author={Meyer, Nico and Ufrecht, Christian and Periyasamy, Maniraman and Plinge, Axel and Mutschler, Christopher and Scherer, Daniel D. and Maier, Andreas},
  journal={arXiv:2404.06314},
  year={2024},
  doi={10.48550/arXiv.2404.06314}
}

Version History

Initial release (v1.0): April 2024

License

Apache 2.0 License