Oh thanks for sharing! The llama.cpp PRs should generally be fine for now - I'm fixing a few small edge cases as well!