NVIDIA announced that developers can now customize the NVIDIA Nemotron 3 Nano model using Prime Intellect Lab, a platform offering training as a service, to produce a downloadable LoRA adapter. The local setup takes about five minutes. The tutorial demonstrates a complete customization workflow: baseline Nemotron 3 Nano in the math-python environment, training a LoRA adapter with reinforcement learning with verifiable rewards, deploying it, and comparing results. The tutorial uses a standard "hello world" example of Python Math. The environment has a default of 100 turns, but for this experiment, each rollout is capped at five assistant turns. The tutorial also applies to Nemotron 3 Super and Nemotron 3 Ultra.