llama.cpp

76d94392 - move allocation workaround out of ggml-alloc.c