NVIDIA: Nvidia Nemotron Nano 9B V2

NVIDIA Nemotron Nano 9B V2 is a large language model built from the ground up to serve as a unified solution for both reasoning and standard text tasks. Its architecture is distinct, utilizing a hybrid design that primarily combines Mamba-2 and MLP layers with a small set of four attention layers. This configuration is intended to balance efficiency with high-level performance, allowing the model to generate structured reasoning traces before arriving at a final answer. Users can toggle these reasoning capabilities via system prompts, providing flexibility to prioritize speed for simpler tasks or deeper analytical accuracy for complex challenges.

The model was developed using Megatron-LM and NeMo-RL, benefiting from improvements derived from the Qwen lineage to refine its output quality. It supports a diverse range of languages, including English, German, Spanish, French, Italian, and Japanese, and has served as a foundation for further specialized iterations like the Japanese-focused variant. By offering a balance of reasoning depth and operational efficiency, the model is well-suited for commercial applications that require reliable, high-performance language processing in a compact form factor.