Hugging Face spaces implementation seems unable to install latest llama-cpp-python version, so downgrading.

Updated Dockerfile so it actually works. Modified config file to include/exclude models and change relevant options as needed. Fixed thumbs up/down feedback.