A collection of reproducible LLM inference engine benchmarks: SGLang vs. vLLM