NVIDIA-NeMo/Nemotron

Developer Asset Hub for NVIDIA Nemotron — A one-stop resource for training recipes, usage cookbooks, datasets, and full end-to-end reference examples to build with Nemotron models

What it solves

Nemotron provides a family of open, high-efficiency multimodal models and the complete training infrastructure needed to build agentic AI. It addresses the difficulty of reproducing production-grade LLM training by providing full pipelines—from raw data curation and synthetic data generation to supervised fine-tuning (SFT) and reinforcement learning (RL)—rather than isolated examples.

How it works

The project is organized into modular "steps" (reusable units of work) and "recipes" (complete pipelines). It utilizes a hybrid Mamba-Transformer Mixture-of-Experts (MoE) architecture to balance sequence efficiency with reasoning capabilities. The ecosystem integrates with NVIDIA's stack, including Megatron-Bridge for training, NeMo-RL for reinforcement learning, and TensorRT-LLM for optimized deployment.

Who it’s for

It is designed for AI researchers and developers who want to train, fine-tune, or deploy high-performance models for agentic workflows, including those needing specialized capabilities in coding, math, scientific reasoning, and multimodal perception (text, image, video, and audio).

Highlights

  • Comprehensive Model Tiers: Offers models ranging from Nano (edge/PC) to Ultra (datacenter-scale), including the 550B-A55B Ultra model.
  • Full Lifecycle Tooling: Includes a CLI for modular steps covering data curation, synthetic data generation (SDG), and benchmark evaluation.
  • Multimodal Capabilities: The Nano Omni model provides native support for text, image, video, and audio in a single decoder.
  • Advanced Training Techniques: Implements Multi-Token Prediction (MTP), asynchronous GRPO, and progressive context scaling up to 1M tokens.

Related

  • Dispatch
    NVIDIA Nemotron 3 Nano Omni Support in vLLM

    vLLM now supports NVIDIA Nemotron 3 Nano Omni, a highly efficient 30B MoE multimodal model that unifies vision, audio, and language reasoning in a single loop to power agentic AI.

  • Dispatch
    NVIDIA Nemotron 3 Ultra Support on vLLM

    vLLM announces Day-0 support for NVIDIA Nemotron 3 Ultra, a 550B parameter model optimized for long-running autonomous agent workflows through a hybrid Transformer-Mamba MoE architecture.

  • Dispatch
    NVIDIA Nemotron 3 Nano Omni release notes / what's new

    NVIDIA has released Nemotron 3 Nano Omni, an omni-modal model capable of long-context reasoning across text, images, video, and audio, delivering best-in-class accuracy on document intelligence and video understanding benchmarks.

  • Dispatch
    NVIDIA Nemotron 3 Ultra Release

    NVIDIA Nemotron 3 Ultra is a 550B parameter open model optimized for long-running agentic workflows, featuring a 1M token context window and NVFP4 quantization for high efficiency.

  • Dispatch
    NVIDIA Nemotron 3 Nano open evaluation recipe with NeMo Evaluator

    NVIDIA released the 30B Nemotron 3 Nano model with a fully open evaluation recipe built on NeMo Evaluator, enabling anyone to reproduce its benchmark scores and audit the entire evaluation pipeline.