A service must expose several trained models through a managed inference-serving process. Which NVIDIA component fits?

1. A GPU device plugin alone
2. Triton Inference Server
3. A model training optimizer alone
4. The NVIDIA driver alone