llama.cpp

Q4_2 quantization with rmse-optimized scale and quants

#1062

Merged