https://news.ycombinator.com/item?id=48700782
Will propose a patch back to llama.cpp or provide it as a fork.