nemotron

NVIDIA Nemotron Skill

Family of open LLMs with hybrid Mamba-Transformer MoE architecture, 1M-token context windows, optimized for agentic AI with NIM deployment and synthetic data generation.

Official Sources:

• NVIDIA Nemotron
• Nemotron 3 Research
• NeMo Framework Docs
• NIM for LLMs
• Synthetic Data Generation

Model Family

Reasoning Models

# Nano 30B A3B - Cost-efficient agentic tasks
# - 3.2B active params (31.6B total)
# - 1M token context
# - 4x faster than Nemotron 2 Nano

# Super 49B - Multi-agent reasoning
# - ~49B params
# - High accuracy, efficient deep research

# Ultra 253B - Enterprise workflows
# - ~253B params
# - Maximum accuracy for complex scenarios

Specialized Models

# Vision Language 12B
# - Document intelligence
# - Video understanding

# RAG Models
# - Extraction, embedding, reranking

# Safety Models
# - Jailbreak detection
# - Content safety (multilingual)

# Speech Models
# - ASR, TTS, neural MT

Quick Start with NIM

# Get NGC API key from https://build.nvidia.com

# Authenticate Docker
echo "$NGC_API_KEY" | docker login nvcr.io --username '$oauthtoken' --password-stdin

# Deploy Nemotron Nano
docker run -it --rm --gpus all \
  -e NGC_API_KEY=$NGC_API_KEY \
  -p 8000:8000 \
  nvcr.io/nim/nvidia/nemotron-3-nano-30b-a3b:1.0.0

# Wait for startup
# INFO: Application startup complete
# INFO: Uvicorn running on http://0.0.0.0:8000

API Usage

from openai import OpenAI

# Connect to NIM
client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="not-used"
)

# Chat completion
response = client.chat.completions.create(
    model="nvidia/nemotron-3-nano-30b-a3b",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain quantum computing."}
    ],
    temperature=0.7,
    max_tokens=500
)

print(response.choices[0].message.content)

# Streaming
for chunk in client.chat.completions.create(
    model="nvidia/nemotron-3-nano-30b-a3b",
    messages=[{"role": "user", "content": "Count to 10"}],
    stream=True
):
    print(chunk.choices[0].delta.content or "", end="")