The largest unsloth published gguf also fits and runs just fine on a cpu-only machine with 128GB RAM, using llama-server PR 27742

https://github.com/ggml-org/llama.cpp/pull/27742