Qwen3.6-27B - DASH-Q 2-bit GGUF
llama.cpp GGUF files of Qwen/Qwen3.6-27B 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 |
|---|---|---|---|
Qwen3.6-27B-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 9.33 GB | 2.72 |
Qwen3.6-27B-DASHQ-IQ2_XS.gguf |
IQ2_XS | 10.04 GB | 2.93 |
Qwen3.6-27B-DASHQ-IQ2_M.gguf |
IQ2_M | 10.65 GB | 3.11 |
Qwen3.6-27B-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 11.37 GB | 3.32 |
Text only: the vision tower is not included.
Perplexity (lower is better)
| Type | Model | Size | WikiText-2 | C4 |
|---|---|---|---|---|
| F16 | 54.7 GB | 6.25 | 9.85 | |
| IQ2_XXS | llama.cpp IQ2_XXS (imatrix) | 8.67 GB | 7.91 | 12.03 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 9.39 GB | 8.15 | 11.35 |
| IQ2_XXS | DASH-Q IQ2_XXS | 9.33 GB | 7.78 | 11.24 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 9.33 GB | 7.67 | 11.35 |
| IQ2_XS | DASH-Q IQ2_XS | 10.04 GB | 7.00 | 10.65 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 10.24 GB | 7.01 | 10.85 |
| IQ2_M | unsloth UD-IQ2_M | 10.85 GB | 6.85 | 10.48 |
| IQ2_M | DASH-Q IQ2_M | 10.65 GB | 6.59 | 10.33 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 10.86 GB | 6.67 | 10.66 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 11.85 GB | 6.83 | 10.38 |
| Q2_K_XL | DASH-Q Q2_K_XL | 11.37 GB | 6.61 | 10.27 |
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 Q2_K_XL | 79.7 | 61.4 | 81.9 | 81.8 | 77.0 | 44.6 | 85.2 | 55.2 | 75.5 | 71.37 |
Usage
llama-cli -m Qwen3.6-27B-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (Qwen/Qwen3.6-27B).