Perceptrain is a Python package that provides a simple interface to execute distributed machine learning training. It supports customization, gradient-based, gradient-free optimizations and various experiment tracking methods.

**For more detailed information, check out the documentation.

**For any questions or comments, feel free to start a discussion. **

Feature highlights

• Training models made simple with Trainer and Train Configurations.

• Support for gradient based and gradient free optimization.

• Accelerator supported distributed training made simple across multi node/multi gpu setups.

• Extensive callbacks, along with support for mlflow and tensorboard tracking.

Installation guide

perceptrain is available on PyPI and can be installed using pip as follows:

pip install perceptrain

The default, pre-installed backend for perceptrain is PyQTorch, a differentiable state vector simulator. It is possible to install additional following extras:

• mlflow: For experiment tracking.

To install individual extras, use the following syntax (IMPORTANT Make sure to use quotes):

pip install "perceptrain[mlflow]"

To install all available extras, simply do:

pip install "perceptrain[all]"

Contributing

Before making a contribution, please review our code of conduct.

• Submitting Issues: To submit bug reports or feature requests, please use our issue tracker.
• Developing in perceptrain: To learn more about how to develop within perceptrain, please refer to contributing guidelines.

Setting up perceptrain in development mode

We recommend to use the hatch environment manager to install perceptrain from source:

python -m pip install hatch

# get into a shell with all the dependencies
python -m hatch shell

# run a command within the virtual environment with all the dependencies
python -m hatch run python my_script.py

WARNING hatch will not combine nicely with other environment managers such as Conda. If you still want to use Conda, install it from source using pip:

# within the Conda environment
python -m pip install -e .

Users also report problems running Hatch on Windows, we suggest using WSL2.

Citation

If you use perceptrain for a publication, we kindly ask you to cite our work using the following BibTex entry:

@article{perceptrain2024pasqal,
  title = {perceptrain},
  author={Manu Lahariya},
  year = {2025}
}

License

Perceptrain is a free and open source software package, released under the Apache License, Version 2.0.