NeMo RL

NeMo RL is NVIDIA's open-source post-training library for streamlining and scaling reinforcement-learning methods for multimodal models such as LLMs and VLMs, from small experiments to multi-GPU, multi-node deployments.