unsloth
NVIDIA Nemotron 3 Nano 4B
NVIDIA Nemotron 3 Nano 4B is a 3.97B dense nemotron_h model from unsloth. Configured context length is 1,048,576 tokens. License: other. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters4.0B
Context1024K
Attentiongqa
Licenseother
42 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 7.41 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 5.24 GB | Hugging Face file bytes |
| Q8_0 | 3.94 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 4.25 GB | Hugging Face file bytes |
| Q6_K | 3.78 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 3.08 GB | Hugging Face file bytes |
| Q5_K_M | 2.94 GB | Hugging Face file bytes |
| Q5_K_S | 2.90 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 2.92 GB | Hugging Face file bytes |
| Q4_K_M | 2.70 GB | Hugging Face file bytes |
| Q4_K_S | 2.64 GB | Hugging Face file bytes |
| IQ4_NL | 2.39 GB | Hugging Face file bytes |
| IQ4_XS | 2.37 GB | Hugging Face file bytes |
| Q4_1 | 2.52 GB | Hugging Face file bytes |
| Q4_0 | 2.35 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 2.50 GB | Hugging Face file bytes |
| Q3_K_M | 2.30 GB | Hugging Face file bytes |
| Q3_K_S | 2.19 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 2.23 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 2.33 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 2.03 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Tight fit | UD-Q8_K_XL | 23–31 |
| GeForce RTX 3060 12GB | Runs | BF16 | 21–29 |
| GeForce RTX 4060 Ti 16GB | Runs | BF16 | 17–23 |
| GeForce RTX 4090 | Runs great | BF16 | 59–81 |
| GeForce RTX 5090 | Runs great | BF16 | 105–143 |
| Apple M4 Pro | Runs great | BF16 | 16–22 |
| Apple M5 | Runs | BF16 | 9.0–12 |
| CPU only · DDR5-5600 | Runs on CPU | BF16 | 5.3–7.2 |
Smallest devices that fit it fully
- GeForce RTX 3050 6GB — Tight fit, Q8_0, 19–26 tok/s
- GeForce RTX 2060 6GB — Tight fit, Q8_0, 37–51 tok/s
- GeForce GTX 1660 Super — Tight fit, Q8_0, 37–51 tok/s
- GeForce GTX 1660 Ti — Tight fit, Q8_0, 32–44 tok/s
- Intel Arc A380 — Tight fit, Q8_0, 21–28 tok/s
- GeForce RTX 5060 Ti 8GB — Tight fit, UD-Q8_K_XL, 35–48 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (BF16).
llama.cpp
llama-server -hf unsloth/NVIDIA-Nemotron-3-Nano-4B-GGUF:BF16 -c 8192 -ngl 42
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/NVIDIA-Nemotron-3-Nano-4B-GGUF # Then load quant BF16