NVIDIA Unveils Nemotron 3 Super: An Open-Source Hybrid-Architecture Model Tailored for Sophisticated Inference Tasks
3 day ago / Read about 0 minute
Author:小编   

NVIDIA has officially launched Nemotron 3 Super, an open-source large language model meticulously crafted for intricate multi-agent reasoning scenarios. This innovative model integrates a hybrid Mamba-Transformer architecture, enhanced by a Mixture of Experts (MoE) mechanism. It boasts an expansive context window capable of accommodating up to one million tokens, while delivering a remarkable fourfold surge in inference speed when compared to its forerunner. Notably, NVIDIA has made the complete model weights, datasets, and deployment solutions freely accessible to the public.