Ministral-3-3B-Reasoning-2512 - DASH-Q 2-bit GGUF
llama.cpp GGUF files of mistralai/Ministral-3-3B-Reasoning-2512 quantized with DASH-Q, at several 2-bit-class sizes. Every file uses only standard llama.cpp tensor types (no tensor above 4 bits) and loads in any recent llama.cpp build.
| File | Type | Size | Bits / weight |
|---|---|---|---|
Ministral-3-3B-Reasoning-2512-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 1.06 GB | 2.46 |
Ministral-3-3B-Reasoning-2512-DASHQ-IQ2_XS.gguf |
IQ2_XS | 1.22 GB | 2.85 |
Ministral-3-3B-Reasoning-2512-DASHQ-IQ2_M.gguf |
IQ2_M | 1.29 GB | 3.00 |
Ministral-3-3B-Reasoning-2512-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 1.39 GB | 3.24 |
Text only: the vision tower is not included.
Perplexity (lower is better)
| Type | Model | Size | WikiText-2 | C4 |
|---|---|---|---|---|
| IQ2_XXS | llama.cpp IQ2_XXS (imatrix) | 1.10 GB | 14.18 | 24.68 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 1.11 GB | 14.49 | 24.42 |
| IQ2_XXS | DASH-Q IQ2_XXS | 1.06 GB | 13.15 | 22.33 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 1.19 GB | 12.03 | 20.70 |
| IQ2_XS | DASH-Q IQ2_XS | 1.22 GB | 10.44 | 18.21 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 1.32 GB | 10.16 | 17.53 |
| IQ2_M | unsloth UD-IQ2_M | 1.34 GB | 10.22 | 17.51 |
| IQ2_M | DASH-Q IQ2_M | 1.29 GB | 9.86 | 17.02 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 1.46 GB | 10.26 | 17.81 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 1.49 GB | 9.89 | 16.95 |
| Q2_K_XL | DASH-Q Q2_K_XL | 1.39 GB | 9.51 | 16.51 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Usage
llama-cli -m Ministral-3-3B-Reasoning-2512-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (mistralai/Ministral-3-3B-Reasoning-2512).