Nemotron-4 Models by NVIDIA AI

About

The Nemotron-4 340B family consists of large language models (LLMs) that are openly accessible and tailored for synthetic data generation, crucial for training other LLMs 34. This innovative suite includes a base model, instruct model, and reward model, each serving unique purposes. The base model, trained on an extensive 9 trillion token dataset, supports the instruct model in producing diverse synthetic data that emulates real-world scenarios, while the reward model focuses on evaluating and refining outputs for helpfulness and coherence 12. Optimized for NVIDIA's NeMo framework and TensorRT-LLM for inference, these models are designed for both research and commercial use due to their open licensing. Moreover, the fully open-sourced pipeline encourages AI community collaboration and innovation 9.

Specifications(3 models)

Available From(2 providers)

DeepInfra

NVIDIA NIM

Pricing

Related Model Families

NVIDIA Nemotron Nano 12B v2 VL

NemoGuard

NV-EmbedNemotron-Labs-DiffusionNemotron-Labs TwoTower

Nemotron 3

Models(3)