amagi-tokbench

One command to benchmark a local LLM server (Ollama, llama.cpp, vLLM): prompt and generation tokens/sec, time to first token, peak VRAM, written as a markdown + PNG card you can paste anywhere.