model routing

Practical, bash-first guide to running multiple LLMs concurrently on one Linux box for specialization, A/B tests, and full GPU use. Covers Ollama, llama.cpp, and vLLM setups; per-GPU/CPU pinning, memory caps, and MIG; apt/dnf/zypper installs; starting/stopping via tmux, systemd, or Docker; Nginx routing behind one URL; and monitoring/tuning—so you can reliably serve, isolate, and compare models in parallel.