You lost me on "spare GPU". I don't have any capable GPUs, let alone spare ones :)

This is very promising, definitely looks more user friendly than exo. Can't wait to try it out.

> MoE models via expert sharding with zero cross-node inference traffic

This makes the whole project questionable