thu-pacman/chitu

High-performance inference framework for large language models, focusing on efficiency, flexibility, and availability.