unsloth
NVIDIA Nemotron 3.5 Lightning 30B A3B
NVIDIA Nemotron 3.5 Lightning 30B A3B is a 32.9B mixture-of-experts nemotron_h_moe model from unsloth, with about 3.0B parameters active per token. 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.
Parameters32.9B / 3.0B active
Context1024K
Attentiongqa
Licenseother
52 layers · 2 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 61.3 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 36.0 GB | Hugging Face file bytes |
| Q8_0 | 32.6 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 32.6 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 28.3 GB | Hugging Face file bytes |
| UD-Q5_K_M | 28.1 GB | Hugging Face file bytes |
| UD-Q5_K_S | 24.4 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 23.8 GB | Hugging Face file bytes |
| UD-Q4_K_M | 23.5 GB | Hugging Face file bytes |
| UD-Q4_K_S | 22.8 GB | Hugging Face file bytes |
| IQ4_NL | 19.8 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 19.8 GB | Hugging Face file bytes |
| UD-IQ3_S | 19.7 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 18.4 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 18.1 GB | Hugging Face file bytes |
| UD-IQ1_M | 18.1 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Expert offload | UD-Q4_K_M | 58–79 |
| GeForce RTX 3060 12GB | Expert offload | UD-Q4_K_M | 72–99 |
| GeForce RTX 4060 Ti 16GB | Expert offload | UD-Q4_K_M | 58–79 |
| GeForce RTX 4090 | Tight fit | IQ4_NL | 234–319 |
| GeForce RTX 5090 | Tight fit | UD-Q5_K_XL | 307–419 |
| Apple M4 Pro | Won't fit | UD-IQ2_XXS | — |
| Apple M5 | Won't fit | UD-IQ2_XXS | — |
| CPU only · DDR5-5600 | Runs on CPU | UD-Q4_K_XL | 19–26 |
Smallest devices that fit it fully
- GeForce RTX 4090 — Tight fit, IQ4_NL, 234–319 tok/s
- GeForce RTX 3090 Ti — Tight fit, IQ4_NL, 234–319 tok/s
- GeForce RTX 3090 — Tight fit, IQ4_NL, 217–296 tok/s
- NVIDIA L4 — Tight fit, IQ4_NL, 70–95 tok/s
- NVIDIA A10 — Tight fit, IQ4_NL, 139–190 tok/s
- RTX A5000 — Tight fit, IQ4_NL, 178–243 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-Q4_K_M).
llama.cpp
# Expert offload is approximate. llama.cpp can pin experts to CPU: # -ot "\\.ffn_.*_exps\\.=CPU" llama-server -hf unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:UD-Q4_K_M -c 8192 -ngl 99
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF # Then load quant UD-Q4_K_M