you can spawn multiple llama.cpp servers and query them simultaneously. It’s actually better this way because you get to run different models for different purposes or do sanity checks via a second model.