the-benchmark

Minimal zero-dependency benchmark for local LLM servers (LM Studio, Ollama, llama.cpp, vLLM): prefill, generation, and a real agentic tool-calling loop.