Lumabri Enables Peer-to-Peer MoE Inference via Colibri Engine

Lumabri runs large Mixture-of-Experts models across a P2P swarm using the dependency-free Colibri engine written in C. The system implements on-demand byte streaming where inference-touched weights are mirrored locally to SSDs for subsequent full-speed execution.