unsloth
Llama 4 Scout 17B 16E Instruct
Llama 4 Scout 17B 16E Instruct is a 107.8B mixture-of-experts llama4 model from unsloth, with about 22.9B parameters active per token. Configured context length is 10,485,760 tokens. License: other. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters108B / 22.9B active
Context10240K
Attentiongqa
Licenseother
48 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 200.8 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 119.4 GB | Hugging Face file bytes |
| Q8_0 | 106.7 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 87.6 GB | Hugging Face file bytes |
| Q6_K | 82.4 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 73.7 GB | Hugging Face file bytes |
| Q5_K_M | 71.3 GB | Hugging Face file bytes |
| Q5_K_S | 69.2 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 57.7 GB | Hugging Face file bytes |
| Q4_K_M | 60.9 GB | Hugging Face file bytes |
| Q4_K_S | 57.2 GB | Hugging Face file bytes |
| IQ4_NL | 56.8 GB | Hugging Face file bytes |
| IQ4_XS | 53.7 GB | Hugging Face file bytes |
| Q4_1 | 62.9 GB | Hugging Face file bytes |
| Q4_0 | 57.0 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 45.7 GB | Hugging Face file bytes |
| Q3_K_M | 48.2 GB | Hugging Face file bytes |
| Q3_K_S | 43.5 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 42.6 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 39.5 GB | Hugging Face file bytes |
| Q2_K | 73.9 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 34.8 GB | Hugging Face file bytes |
| UD-IQ1_M | 32.6 GB | Hugging Face file bytes |
| UD-IQ1_S | 30.2 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Partial offload | UD-IQ1_S | 7.9–11 |
| GeForce RTX 3060 12GB | Partial offload | UD-IQ2_XXS | 7.6–10 |
| GeForce RTX 4060 Ti 16GB | Partial offload | UD-IQ2_XXS | 8.1–11 |
| GeForce RTX 4090 | Expert offload | Q3_K_S | 48–65 |
| GeForce RTX 5090 | Expert offload | Q3_K_S | 85–116 |
| Apple M4 Pro | Won't fit | UD-IQ1_S | — |
| Apple M5 | Won't fit | UD-IQ1_S | — |
| CPU only · DDR5-5600 | Won't fit | UD-IQ1_S | — |
Smallest devices that fit it fully
- NVIDIA A100 40GB — Tight fit, UD-IQ2_XXS, 89–122 tok/s
- NVIDIA L40S — Tight fit, UD-IQ3_XXS, 45–62 tok/s
- NVIDIA A40 — Tight fit, UD-IQ3_XXS, 36–50 tok/s
- RTX 6000 Ada — Tight fit, UD-IQ3_XXS, 50–68 tok/s
- RTX A6000 — Tight fit, UD-IQ3_XXS, 40–55 tok/s
- Apple M3 Max (40-core GPU) — Tight fit, UD-IQ1_M, 24–33 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-IQ2_XXS).
llama.cpp
llama-server -hf unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF:UD-IQ2_XXS -c 4096 -ngl 18
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF # Then load quant UD-IQ2_XXS