Ministral-3-8B-Reasoning-2512 - DASH-Q 2-bit GGUF
llama.cpp GGUF files of mistralai/Ministral-3-8B-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-3-8B-Reasoning-2512-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 2.64 GB | 2.48 |
Ministral-3-8B-Reasoning-2512-DASHQ-IQ2_XS.gguf |
IQ2_XS | 2.99 GB | 2.81 |
Ministral-3-8B-Reasoning-2512-DASHQ-IQ2_M.gguf |
IQ2_M | 3.17 GB | 2.97 |
Ministral-3-8B-Reasoning-2512-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 3.44 GB | 3.23 |
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) | 2.53 GB | 9.89 | 16.96 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 2.63 GB | 9.97 | 16.55 |
| IQ2_XXS | DASH-Q IQ2_XXS | 2.64 GB | 9.36 | 16.18 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 2.75 GB | 8.79 | 15.06 |
| IQ2_XS | DASH-Q IQ2_XS | 2.99 GB | 8.15 | 14.12 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 3.11 GB | 7.73 | 13.37 |
| IQ2_M | unsloth UD-IQ2_M | 3.16 GB | 7.74 | 13.28 |
| IQ2_M | DASH-Q IQ2_M | 3.17 GB | 7.59 | 13.18 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 3.35 GB | 7.81 | 13.45 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 3.57 GB | 7.62 | 13.12 |
| Q2_K_XL | DASH-Q Q2_K_XL | 3.44 GB | 7.38 | 12.82 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Usage
llama-cli -m Ministral-3-8B-Reasoning-2512-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (mistralai/Ministral-3-8B-Reasoning-2512).