Fetching the paper…

Helix: Serving Large Language Models over Heterogeneous GPUs and Network via Max-Flow · Around