Colibri

Colibri is a lightweight engine designed to run the GLM-5.2 model, featuring 744 billion parameters, on consumer machines with just 25GB of RAM. This innovative tool addresses the challenge of running large-scale models without the need for complex dependencies, making it accessible for developers and researchers. Built in pure C, Colibri streams expert models directly from disk, ensuring efficient memory usage and performance. It's perfect for AI enthusiasts, developers, and researchers looking to leverage advanced machine learning capabilities without requiring high-end hardware.