Qwen3.5-9B - DASH-Q 2-bit GGUF
llama.cpp GGUF files of Qwen/Qwen3.5-9B 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 |
|---|---|---|---|
Qwen3.5-9B-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 3.20 GB | 2.78 |
Qwen3.5-9B-DASHQ-IQ2_XS.gguf |
IQ2_XS | 3.42 GB | 2.97 |
Qwen3.5-9B-DASHQ-IQ2_M.gguf |
IQ2_M | 3.64 GB | 3.16 |
Qwen3.5-9B-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 3.93 GB | 3.42 |
Text only: the vision tower is not included.
Perplexity (lower is better)
| Type | Model | Size | WikiText-2 | C4 |
|---|---|---|---|---|
| F16 | 18.4 GB | 7.07 | 11.17 | |
| IQ2_XXS | llama.cpp IQ2_XXS (imatrix) | 3.24 GB | 9.94 | 15.87 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 3.19 GB | 9.54 | 15.04 |
| IQ2_XXS | DASH-Q IQ2_XXS | 3.20 GB | 9.50 | 14.93 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 3.42 GB | 9.07 | 14.19 |
| IQ2_XS | DASH-Q IQ2_XS | 3.42 GB | 8.44 | 13.24 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 3.74 GB | 8.02 | 12.92 |
| IQ2_M | unsloth UD-IQ2_M | 3.65 GB | 8.00 | 12.66 |
| IQ2_M | DASH-Q IQ2_M | 3.64 GB | 8.02 | 12.50 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 3.91 GB | 8.06 | 13.10 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 4.12 GB | 8.18 | 12.52 |
| Q2_K_XL | DASH-Q Q2_K_XL | 3.84 GB | 7.68 | 12.28 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Zero-shot accuracy (lm-eval-harness, 0-shot)
| File | arc_easy | arc_challenge | piqa | hellaswag | winogrande | openbookqa | commonsense_qa | truthfulqa_mc2 | lambada_openai | avg |
|---|---|---|---|---|---|---|---|---|---|---|
| DASH-Q IQ2_XXS | 76.4 | 48.4 | 76.2 | 65.4 | 68.0 | 38.4 | 73.7 | 46.5 | 63.1 | 61.77 |
| DASH-Q IQ2_XS | 78.8 | 52.0 | 77.5 | 70.7 | 69.1 | 41.2 | 76.7 | 47.2 | 67.8 | 64.56 |
| DASH-Q IQ2_M | 74.9 | 53.0 | 78.7 | 73.2 | 71.0 | 41.2 | 80.2 | 45.6 | 69.1 | 65.21 |
| DASH-Q Q2_K_XL | 79.0 | 55.9 | 79.4 | 73.9 | 70.9 | 43.0 | 78.8 | 48.9 | 69.6 | 66.60 |
Usage
llama-cli -m Qwen3.5-9B-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (Qwen/Qwen3.5-9B).