tokens-per-second
Benchmark local LLM inference engines (llama.cpp, omlx) on your own hardware with real tokens/sec, not borrowed numbers