Benchmark local LLM inference speed (tokens/sec) on your own hardware — llama.cpp native…