In CPU only setups the LibOllamaPath was omitted causing us not to load the ggml-cpu-XXX libraries during inference.