NVIDIA-NeMo/Nemotron
NVIDIA NEMOTRON: TRAINING RECIPES AND DEPLOYMENT
NVIDIA repository with training recipes, deployment guides, and usage examples for Nemotron models. Supports multimodal agents with up to 1M token context.
// KEY FEATURES
- Ready training recipes for Nemotron 3.5 Super, Lightning, Ultra, Nano Omni
- Modular steps for data prep, training, evaluation, and SDG
- Claude Code plugin nemotron-customize for pipeline building
- Deployment on edge, GPU, or data center via NIM microservices
// DEEP DIVE: HOW IT WORKS
The repository is organized into four sections: modular steps for individual lifecycle stages, complete training recipes from raw data to model, usage cookbooks for deployment and inference, and use-case examples for agentic workflows. The Claude Code plugin nemotron-customize reads the step catalog from src/nemotron/steps/ and generates YAML pipeline configs, validating artifact dependencies.
Unlike simple model hubs, this publishes full reproducible training pipelines: configs, scripts, quality evaluation. The Nemotron family includes hybrid Mamba-Transformer MoE architectures with Multi-Token Prediction: 3.5 Super (120B-A12B, 1M context, video/images), 3.5 Lightning (30B-A3B, high throughput), 3 Ultra, and 3 Nano Omni (audio/video/text). Optimization via TensorRT-LLM and NIM enables flexible deployment.
The Claude Code plugin requires running from the repository root (where pyproject.toml and src/nemotron/steps/ exist). Training recipes target NVIDIA infrastructure (TensorRT-LLM, NIM). Model weights are hosted on Hugging Face; the repository contains only pipeline configurations and code.
// USE CASES
Best for: For ML engineers and researchers to train and deploy agentic models
- Reproduce full Nemotron 3.5 Super training from data to weights
- Deploy multimodal agent with video and audio support via NIM
- Build custom training pipeline via nemotron-customize plugin in Claude Code
// QUICK START
/plugin marketplace add NVIDIA-NeMo/Nemotron
/plugin install nemotron-customize@nvidia-nemotron
cd /path/to/Nemotron
claude
/nemotron-customizeCommands are taken from the repository README — check the current version on GitHub before running.
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram