NVIDIA post-trains Nemotron 3.5 Super VL with RL
Nvidia details a practical recipe for post-training its Nemotron 3.5 Super VL model for scientific applications using reinforcement learning and teacher-model distillation. The workflow maps experimental records into verifiable prompts, uses NeMo Gym and NeMo RL for training, and measures success via Tanimoto molecular similarity scoring. Open article →