NVIDIA NeMo-RL is a new open source post-training library for reinforcement learning that scales from single-GPU prototypes to thousand-GPU deployments. The library features native Hugging Face integration, optimized training and inference, popular algorithms like DPO and GRPO, and Ray-based orchestration. A practical
Table of contents
Training high-performing reasoning models with NeMo-RLStep-by-step training processResultsGet started with NeMo-RLSort: