Slight quantization improvement for Q4_K and Q5_K#5361

ikawrakow merged 2 commits intomasterggml-org/llama.cpp:masterfrom

ik/q4k_tuningggml-org/llama.cpp:ik/q4k_tuningCopy head branch name to clipboard