Full threadbeiller·Here is a thread to tweak the parameters which the model seems very sensitive to:https://github.com/ggerganov/llama.cpp/issues/129View on HN