PyKaruiFlow is a university project. In this project we are trying to imitate the famous PyTorch framework for educational purposes and find an answer to the following questions:
- how does computational graph work?
- how does backpropagation work?
- how PyTorch does everything dynamically compared to TensorFlow 1.x?
This is a Python version of the upcoming KaruiFlow implementation with CUDA. The goal of this Python version is to determine the API and overall functionality as well as figure out the theory behind computational graphs and backpropagation.
This implementation surely won't be as performant or flexible as PyTorch, but it won't contain any boilerplate (or autogenerated) code or be as massive. Although some technical effort will be put into designing KaruiFlow, still it must be great for educational purposes and help a lot to understand how different frameworks work under the hood.
- pip install --upgrade git+https://github.com/oKatanaaa/PyKaruiFlow.git