What is NVIDIA and what is it used for?
In AI, NVIDIA offers the open Nemotron model family, the NeMo suite for preparing data, training, customizing, evaluating and governing models and agents, and NIM microservices for deploying them.
NVIDIA describes Nemotron as a family of multimodal models for agentic reasoning. The Nemotron 3 generation uses a hybrid Mamba-Transformer mixture-of-experts (MoE) architecture and comes in Nano, Super and Ultra sizes, according to NVIDIA's model cards on Hugging Face.
The weights are published on Hugging Face under the "nvidia" organization. The models are also available as NIM microservices, which NVIDIA presents as part of NVIDIA AI Enterprise. At build.nvidia.com you can try the models and download the NIM packages.
NeMo bundles components that work together:
- NeMo Customizer: fine-tuning and reinforcement learning with your own data.
- NeMo Retriever: extraction of multimodal data from complex documents, with Nemotron RAG models.
- NeMo Guardrails: a programmable runtime layer for safety and topical relevance, with Nemotron Safety models.
- NeMo Curator: an open source library for curating multimodal datasets.
NVIDIA states that NeMo works with Nemotron and with other models.
