NVIDIA · 30B (3B active) · Mixture of Experts
MoE with 1M context and 3B active
Use Cases
Mixture of Experts
| Quant | Bits | VRAM | Quality | Status |
|---|---|---|---|---|
| Q2_K | 2 | 10.1 GB | low | — |
| Q3_K_M | 3 | 13.9 GB | moderate | — |
| Q4_K_M | 4 | 15.9 GB | good | — |
| Q5_K_M | 5 | 19.7 GB | good | — |
| Q6_K | 6 | 23.6 GB | excellent | — |
| Q8_0 | 8 | 31.2 GB | excellent | — |
| F16 | 16 | 62 GB | lossless | — |
GPUs that run Nemotron 3 Nano 30B
Nemotron 3 Nano 30B needs about 15.9 GB of VRAM at Q4_K_M, so it runs comfortably on a GPU with 20 GB of VRAM or more (39 of the GPUs we track). Common cards, smallest first:
| VRAM | GPUs |
|---|---|
| 20 GB | RTX 4000 SFF Ada , RX 7900 XT , RTX 4000 Ada |
| 24 GB | RTX 4090 , RX 7900 XTX , RTX 3090 Ti |
| 32 GB | RTX 5090 , RTX 5000 Ada , Tesla V100 32GB |
About this model
Nemotron-3-Nano
Nemotron 3 Nano 4B
ollama run nemotron-3-nano:4b
Nemotron 3 Nano 30B
ollama run nemotron-3-nano:30b
Ollama’s Cloud
ollama run nemotron-3-nano:30b-cloud
What is Nemotron?
NVIDIA Nemotron™ is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents.
Nemotron 3 Nano is a large language model (LLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and non-reasoning tasks. It responds to user queries and tasks by first generating a reasoning trace and then concluding with a final response. The model’s reasoning capabilities can be configured through a flag in the chat template. If the user prefers the model to provide its final answer without intermediate reasoning traces, it can be configured to do so, albeit with a slight decrease in accuracy for harder prompts that require reasoning. Conversely, allowing the model to generate reasoning traces first generally results in higher-quality final solutions to queries and tasks.
The model employs a hybrid Mixture-of-Experts (MoE) architecture, consisting of 23 Mamba-2 and MoE layers, along with 6 Attention layers. Each MoE layer includes 128 experts plus 1 shared expert, with 6 experts activated per token. The model has 3.5B active parameters and 30B parameters in total.
The supported languages include: English, German, Spanish, French, Italian, and Japanese. Improved using Qwen.
License/Terms of Use
Governing Terms: Use of this model is governed by the NVIDIA Open Model License Agreement.