Reading the “paper”, this is just a job runner built on Ollama. Exo, and the teams at Apple that built the foundations Exo uses, actually use modern distributed compute.

I don't get why they didn't use llama.cpp instead.