Skip to main content
Ctrl+K
Nemotron - Home Nemotron - Home

Nemotron

  • GitHub
Nemotron - Home Nemotron - Home

Nemotron
  • GitHub

Table of Contents

Nemotron

  • Home
  • Application Examples
  • Deployment Guides
  • Adapting Nemotron to a New Language

Nemotron Step Basics

  • About
  • Basics
  • Getting Started
  • Airgap Environment

Data Curation

  • About
  • Getting Started
  • Concepts
    • Two-Run Curation: Measure, Approve, Apply
    • Signals and Language Packs
    • Curation Pipeline Artifacts
  • Tasks
    • Run the Measure and Apply Flow
    • Apply a Known Policy
    • Evaluate a Policy
    • Audit a Curated Corpus
    • Decontaminate and Subset a Corpus
    • Run Curation on Local JSONL
    • Use a Hugging Face Snapshot
    • Enable Curation Filters
  • Reference
    • Curation CLI
    • Curation Input and Output Format
    • Curation Troubleshooting
    • curate/ingest
    • curate/profile
    • curate/nemo_curator
    • curate/audit
    • curate/decontamination
    • curate/subset
    • Curate Flow Configuration
    • Filter Policy File

Synthetic Data Generation

  • About
  • Getting Started
  • Tips for Using Agents
  • Planning
  • Tasks
    • Create a Domain Dataset
    • Create Tool-Calling Dataset
    • Generate Preference Data for DPO
    • Generate Persona MCQ Data
    • Dispatch SDG to a Cluster
    • Tips for the Data Generation Pipeline
  • Reference
    • Config Schema
    • CLI Reference
    • Output Projections
    • Persona MCQ Configuration Reference
    • Troubleshooting

Translation

  • About
  • Getting Started
  • Tips for Using Agents
  • Concepts
    • Pipeline Overview
    • Segmentation
    • FAITH Evaluation Inside Translation
  • Tasks
    • Run LLM Translation
    • Run NMT Translation
    • Run Google or AWS Translation
    • Configure Fields and Output
    • Use Fine Segmentation
    • Run FAITH Evaluation
  • Reference
    • Translation YAML Reference
    • CLI Reference for Translation
    • Input and Output Format
    • Troubleshooting

Build MCQ Benchmarks

  • About
  • Getting Started
  • Concepts
    • Pipeline Overview
    • Data Preparation
    • Get the Right Questions
    • Question Generation
    • Quality Validation
    • Easiness and Hallucination Filtering
    • Translation
  • Tasks
    • Prepare Data
    • Use Your Domain Data
    • Model Endpoints
    • Tune Prompts
    • Skip Stages
  • Reference
    • Supported Hugging Face Benchmarks
    • Output Files
    • Generation Configuration Reference
    • Translation Configuration Reference
    • Troubleshooting

Build Function-Calling Benchmarks

  • About
  • Tutorials
    • Build Your First Benchmark
    • Run Assisted Authoring End to End
    • Publish and Evaluate a Manual Pack
  • Concepts
    • Pipeline Overview
    • Pipeline Worked Example
    • The Oracle Pack
    • Authoring Flows
    • Evaluation
  • Tasks
    • From Domain Assets To An Oracle Pack
    • Author a Pack
    • Assisted Authoring
    • Onboard an MCP Server
    • Publish a Release
    • Translate a Release
    • Run an Evaluation
  • Reference
    • Oracle Pack Inputs
    • Domain Brief
    • Data Designer Provider Configuration
    • Probe Plan
    • Manifest
    • Tool Catalog And Fixtures
    • Python Backend
    • Task Templates
    • Assertions
    • Validation Cases
    • Endpoint Configuration
    • Held-Out Policy
    • Generation Config
    • Evaluation Config
    • Output Files
    • Troubleshooting

Tokenizer Extension

  • About
  • Prerequisites
  • Getting Started
  • Concepts
    • Pipeline Overview
    • Add, Replace, and Expand
    • Embedding Initialization
    • Evaluation Metrics
  • Tasks
    • Extend a Tokenizer
    • Initialize Embeddings
    • Evaluate a Tokenizer
    • Add a Language
    • Run on a Cluster
  • Reference
    • Extend YAML Reference
    • Init Embeddings YAML Reference
    • Evaluate YAML Reference
    • Eval Init YAML Reference
    • CLI Reference
    • Output Files
    • Troubleshooting

Model Training

  • About
  • Post-training Calculator
  • Getting Started
  • Tips for Using Agents
  • Concepts
    • Training Basics
    • Artifact Graph
    • Training Libraries
  • Tasks
    • Run SFT with AutoModel on Custom Data
    • Choose an SFT Backend
    • Choose a PEFT Backend
    • Choose an RL Alignment Step
    • Run Post-Training Optimization
    • Environment Profiles and Executors
    • Data and Checkpoint Formats
    • Convert Checkpoints
  • Reference
    • Nemotron Steps CLI Reference
    • Env Profile Generator
    • Step Catalog
    • Configuration Conventions
    • SFT Steps
      • sft/automodel
      • sft/megatron_bridge
    • PEFT Steps
      • peft/automodel
      • peft/megatron_bridge
    • RL Steps
      • rl/nemo_rl/dpo
      • rl/nemo_rl/rlvr
      • rl/nemo_rl/rlhf
    • Optimization Steps
      • optimize/modelopt/quantize
      • optimize/modelopt/prune
      • optimize/modelopt/distill
    • Checkpoint Conversion Steps
      • convert/hf_to_megatron
      • convert/megatron_to_hf
      • convert/merge_lora
  • FAQ

Model Evaluation

  • About
  • Getting Started
  • Tips for Using Agents
  • Concepts
    • Pipeline Overview
    • Endpoint Types And Task Families
    • Tokenizer Alignment
  • Tasks
    • Discover The Model Evaluation Step
    • Run A Direct-Mode Evaluation
    • Run A Benchmark Suite
    • Run A Hosted Evaluation
    • Evaluate A Deployed Checkpoint
  • Reference
    • Configuration Reference
    • CLI Reference
    • Output Artifacts
    • Tasks Catalog
    • Troubleshooting

Training Recipes

  • Nemotron 3 Nano
    • Stage 0: Pretraining
    • Stage 1: Supervised Fine-Tuning (SFT)
    • Stage 2: Reinforcement Learning (RL)
    • Stage 3: Evaluation
    • Importing Models and Data
  • Nemotron 3.5 Lightning
    • Stage 0: Pretraining
    • Stage 1: Supervised Fine-Tuning (SFT)
    • Stage 2: Reinforcement Learning (RL)
    • Evaluate Nemotron 3.5 Lightning
    • Quantization (PTQ + QAD)
    • Importing Models and Data
  • Nemotron 3 Omni
    • Stage 0: Supervised Fine-Tuning (SFT)
    • Stage 1: Reinforcement Learning (RL)
    • RL Data Preparation
    • Nemotron 3 Nano Omni — Architecture
    • Nemotron 3 Nano Omni — Inference & Deployment
  • Nemotron 3 Super
    • Stage 0: Pretraining
    • Stage 1: Supervised Fine-Tuning (SFT)
    • Stage 2: Reinforcement Learning (RL)
      • Multi-Environment RLVR (Stages 1.1–1.3)
      • SWE-RL (Stages 2.1–2.2)
      • RLHF (Stage 3)
      • RL Data Preparation
    • Stage 4: Evaluation
    • Stage 3: Quantization
  • Nemotron 3.5 Super
    • Stage 0: Pretraining
    • Stage 1: Supervised Fine-Tuning (SFT)
    • Stage 2: Reinforcement Learning (RL)
      • RLVR (Stage 2.1)
      • Specialist Teachers (Stage 2.2)
      • MOPD (Stage 2.3)
    • Stage 3: Evaluation
    • Quantization (PTQ + QAD)
  • Nemotron 3 Ultra
    • Stage 0: Pretraining
    • Stage 1: SFT
    • Nemotron 3 Ultra
    • Quantization
  • Nemotron Embed
    • Multimodal Workflow
    • Text Data Workflow
    • Data Reference
    • SDG Reference
  • Nemotron Rerank
    • Multimodal Workflow
    • Text Data Workflow
  • Artifact Types & Lineage

Nemotron Kit

  • Nemotron Kit
  • NVIDIA AI Stack
  • nemo_runspec Package
  • Execution through NeMo-Run
  • OmegaConf Configuration System
  • Artifact Tracking
  • Weights & Biases Integration
  • CLI Framework
  • Data Preparation Module
  • Xenna Pipeline Observability

Data Recipes

  • Nemotron-CC Data Curation
  • Long-Document SDG

Architecture

  • Nemotron Architecture
  • Design Philosophy
  • CLI Architecture
  • Runspec: [tool.runspec] Specification
  • Curation Concepts

Curation Concepts#

These pages explain the design of the curation category. Read them before choosing thresholds for a new corpus or interpreting an audit report.

  • Two-Run Curation: Measure, Approve, Apply
  • Signals and Language Packs
  • Curation Pipeline Artifacts

previous

Getting Started With Data Curation

next

Two-Run Curation: Measure, Approve, Apply
NVIDIA NVIDIA
Privacy Policy | Your Privacy Choices | Terms of Service | Accessibility | Corporate Policies | Product Security | Contact

Copyright © 2026, NVIDIA Corporation.