Cosmos-RL supports vLLM as the backend of rollout generaton.

And recently, we have added an experimental support for TensorRT-LLM as the backend of rollout generation.