llama.cpp

a8ca18b4 - llama-bench : clarify benchmarked parts of the computation (#16823)