Caten is an experimental deep learning compiler known as Compile+AbstracTENsor. It combines simplicity and flexibility to support a variety of models, like image processing and language generation. The GPT2 demonstration allows for customization with environment variables. The project is currently focused on building a foundational library and welcomes contributions such as feature additions, bug reports, and documentation improvements. The roadmap includes support for various models and formats, training features, and accelerators. To get involved, join the Discord server, check the roadmap, and create a PR. Setting up Caten involves installing Roswell, ISL, and Qlot. Run tests with python, numpy, and pytorch installed.
https://github.com/hikettei/Caten