llama.cpp

d27b3ca1 - ggml : fix repack work size for mul_mat_id (#14292)