Nemotron Nano 2
Open 9B hybrid Mamba-Transformer reasoning model with 128K context, released with a 6.6T-token pretraining dataset.
NVIDA Nemotron Nano 2 is a new hybrid Mamba-Transformer reasoning model that achieves on-par or better accuracies compared to comparably sized leading open models at up to 6x higher throughput. Nemotron Pretraining Dataset v1 is a 6.6 trillion token dataset collection of web crawl, math, code, SFT, and multilingual Q&A data, used to train NVIDIA Nemotron Nano 2.
Official source
NVIDIA Nemotron Nano 2 and the Nemotron Pretraining Dataset v1research.nvidia.com · published