llama.cpp

992cb503 - ggml: allow backend inputs to not create another split (#28387)