Deploy a model behind an endpoint in minutes, or train it federated across a distributed GPU fleet. Your code, your weights, your keys.