llama.cpp

5d804a49 - ggml-backend: raise GGML_MAX_SPLIT_INPUTS (#15722)