Comment by nikwen

It would be great to have upstream llama.cpp support for this!