llama.cpp

f8ad742a - use std::pair instead of ggml_concat