Basically it gives you a read only dashboard that shows the important metrics for your hosts, ray cluster and vllm and access to the vllm log output without login and if you login as an admin it gives you
1. a start/stop/restart with a drop down to choose spark run recipes to start, if you are on a single node it hides the cluster only recipes 2. a HuggingFace “predownload” feature that lets you choose a recipe ad hit a button and it will download the model for that recipe so you are not standing around waiting for it to download after you start the recipe and an option to auto mirror it to your second node 3. an api with token to execute these things in a scripted way and 4. a prometheus compatible endpoint for metrics consumption somewhere else.
