llama.cpp

bdcb8f42 - CUDA: int8 tensor cores for MMQ (q4_K, q5_K, q6_K) (#7860)