logo Welcome to Ravnest’s documentation!

Ravnest is an asynchronous decentralized training framework designed for large modern deep learning models. It leverages the compute power of regular heterogeneous PCs with limited resources connected across the internet to achieve favorable performance metrics. By clustering compute nodes with similar data transfer speeds and computing capabilities, Ravnest removes the requirement for each node to store the entire model, facilitating decentralized training.

Research Paper

GitHub

Warning

Ravnest is in the throes of active development. Brace yourselves for updates! What you see today may not be what you encounter tomorrow. 🌪️

Indices and tables