llama.cpp

27461cd8 - add Q rot when cache is quantized