llama.cpp

c14e0d2f - ggml : always allocate buffers with size multiple of GGML_MEM_ALIGN