llama.cpp

k-quants : fix quantization ranges

#3646

Merged