Qwen3.5-2B - DASH-Q 2-bit GGUF
llama.cpp GGUF files of Qwen/Qwen3.5-2B 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 |
|---|---|---|---|
Qwen3.5-2B-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 0.70 GB | 2.89 |
Qwen3.5-2B-DASHQ-IQ2_XS.gguf |
IQ2_XS | 0.76 GB | 3.11 |
Qwen3.5-2B-DASHQ-IQ2_M.gguf |
IQ2_M | 0.79 GB | 3.23 |
Qwen3.5-2B-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 0.83 GB | 3.40 |
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) | 0.82 GB | 25.67 | 37.89 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 0.77 GB | 22.23 | 30.34 |
| IQ2_XXS | DASH-Q IQ2_XXS | 0.70 GB | 15.44 | 24.17 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 0.86 GB | 19.08 | 27.49 |
| IQ2_XS | DASH-Q IQ2_XS | 0.76 GB | 13.25 | 20.18 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 0.90 GB | 15.24 | 22.94 |
| IQ2_M | unsloth UD-IQ2_M | 0.86 GB | 13.97 | 20.52 |
| IQ2_M | DASH-Q IQ2_M | 0.79 GB | 12.87 | 19.13 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 1.00 GB | 15.39 | 22.75 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 0.97 GB | 13.57 | 19.71 |
| Q2_K_XL | DASH-Q Q2_K_XL | 0.83 GB | 12.02 | 18.69 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Usage
llama-cli -m Qwen3.5-2B-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (Qwen/Qwen3.5-2B).