SGLang vs vLLM: Install, Serve, and Benchmark on Bare Metal

Two open-source engines currently dominate self-hosted LLM inference: vLLM and SGLang. Both promise the exact same thing—feed them a Hugging Face safetensors model, and they will spin up an ultra-fast