🧮 LLM Queue & Throughput

live vLLM /metrics — realtime tok/s (10s sampling) + rolling avgs 15m–24h ·