Nemotron-3-Super Notebooks

A collection of notebooks demonstrating deployment and fine-tuning cookbooks for NVIDIA Nemotron-3-Super.

Overview

These notebooks provide end-to-end recipes for deploying and customizing Nemotron-3-Super.

What’s Inside

Deployment

• vllm_cookbook.ipynb — Deploy Nemotron-3-Super with vLLM.

• sglang_cookbook.ipynb — Deploy Nemotron-3-Super with SGLang.

• trtllm_cookbook.ipynb — Deploy Nemotron-3-Super with TensorRT-LLM.

• SparkDeploymentGuide — DGX Spark single-GPU deployment guide for Nemotron 3 Super with vLLM (nightly) and TensorRT-LLM, including NVFP4 quantization and MTP speculative decoding.

Fine-Tuning

• grpo-dapo — Full-weight RL training with GRPO/DAPO algorithm, reproducing emergent math reasoning from a base model.

• lora-text2sql — Supervised fine-tuning (LoRA) recipe for the Text2SQL use case, including dataset preparation and training with NeMo Megatron-Bridge and NeMo AutoModel libraries.

Agentic

• OpenScaffoldingResources — Guides for using Nemotron-3-Super with agentic coding tools (OpenCode, OpenClaw, Kilo Code CLI, OpenHands) via OpenRouter and build.nvidia.com.