vllm-cli

Why we included this project

Running vLLM by hand means juggling long flag lists, restarting servers, and losing track of what is actually live on the GPU. This CLI gives you two ways around that: a menu-driven terminal with live GPU and system status for day-to-day work, and plain commands that slot into scripts and automation. It also handles chores that normally eat your time, like locating local models already downloaded through HuggingFace or Ollama, keeping ready-made configuration profiles for different hardware, and checking GPU, memory, and CUDA compatibility before you start. A newer experimental feature serves several models behind one API endpoint, waking and sleeping GPUs as demand shifts. For anyone who runs vLLM on a single machine or a small GPU box, it is a thin layer of convenience that makes the engine far less annoying to operate.