llama.cpp

Only use Q6_K for output weights if tensor size is multiple of 256

#1932

Merged