llama.cpp

329112c5 - bring back ggml_concat as it may not work on other backend