llama.cpp

8c583242 - kleidiai: add optimized per-channel kernels for Q8_0 (#16993)