unsloth
Ministral 3 14B Instruct 2512
Ministral 3 14B Instruct 2512 is a 13.5B dense mistral3 model from unsloth. Configured context length is 262,144 tokens. License: apache-2.0. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters13.5B
Context256K
Attentiongqa
Licenseapache-2.0
40 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 25.2 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 15.9 GB | Hugging Face file bytes |
| Q8_0 | 13.4 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 11.3 GB | Hugging Face file bytes |
| Q6_K | 10.3 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 8.98 GB | Hugging Face file bytes |
| Q5_K_M | 8.96 GB | Hugging Face file bytes |
| Q5_K_S | 8.74 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 7.79 GB | Hugging Face file bytes |
| Q4_K_M | 7.67 GB | Hugging Face file bytes |
| Q4_K_S | 7.30 GB | Hugging Face file bytes |
| IQ4_NL | 7.27 GB | Hugging Face file bytes |
| IQ4_XS | 6.92 GB | Hugging Face file bytes |
| Q4_1 | 7.99 GB | Hugging Face file bytes |
| Q4_0 | 7.27 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 6.46 GB | Hugging Face file bytes |
| Q3_K_M | 6.22 GB | Hugging Face file bytes |
| Q3_K_S | 5.66 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 5.12 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 5.15 GB | Hugging Face file bytes |
| Q2_K | 9.92 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 3.78 GB | Hugging Face file bytes |
| UD-IQ1_M | 3.42 GB | Hugging Face file bytes |
| UD-IQ1_S | 3.21 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Tight fit | Q3_K_S | 23–32 |
| GeForce RTX 3060 12GB | Tight fit | UD-Q5_K_XL | 18–25 |
| GeForce RTX 4060 Ti 16GB | Tight fit | UD-Q6_K_XL | 12–16 |
| GeForce RTX 4090 | Runs | UD-Q8_K_XL | 30–41 |
| GeForce RTX 5090 | Tight fit | BF16 | 35–47 |
| Apple M4 Pro | Tight fit | UD-Q8_K_XL | 8.4–12 |
| Apple M5 | Tight fit | UD-Q5_K_XL | 7.7–10 |
| CPU only · DDR5-5600 | Runs on CPU | UD-Q8_K_XL | 2.7–3.6 |
Smallest devices that fit it fully
- GeForce RTX 3050 6GB — Tight fit, UD-IQ2_XXS, 20–27 tok/s
- GeForce RTX 2060 6GB — Tight fit, UD-IQ2_XXS, 39–53 tok/s
- GeForce GTX 1660 Super — Tight fit, UD-IQ2_XXS, 39–53 tok/s
- GeForce GTX 1660 Ti — Tight fit, UD-IQ2_XXS, 33–46 tok/s
- Intel Arc A380 — Tight fit, UD-IQ2_XXS, 22–30 tok/s
- GeForce RTX 5060 Ti 8GB — Tight fit, Q3_K_S, 37–50 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-Q6_K_XL).
llama.cpp
llama-server -hf unsloth/Ministral-3-14B-Instruct-2512-GGUF:UD-Q6_K_XL -c 8192 -ngl 40
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/Ministral-3-14B-Instruct-2512-GGUF # Then load quant UD-Q6_K_XL