llama.cpp

5d9e5997 - ggml : adapt AMX to tensor->grad removal (#0)