Qwen2.5-VL-32B-Instruct - DASH-Q 2-bit GGUF
llama.cpp GGUF files of Qwen/Qwen2.5-VL-32B-Instruct 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 |
|---|---|---|---|
Qwen2.5-VL-32B-Instruct-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 9.31 GB | 2.27 |
Qwen2.5-VL-32B-Instruct-DASHQ-IQ2_XS.gguf |
IQ2_XS | 10.68 GB | 2.61 |
Qwen2.5-VL-32B-Instruct-DASHQ-IQ2_M.gguf |
IQ2_M | 11.47 GB | 2.80 |
Qwen2.5-VL-32B-Instruct-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 12.62 GB | 3.08 |
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) | 9.03 GB | 7.78 | 12.27 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 9.30 GB | 7.82 | 12.23 |
| IQ2_XXS | DASH-Q IQ2_XXS | 9.31 GB | 8.11 | 11.98 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 9.96 GB | 7.24 | 11.52 |
| IQ2_XS | DASH-Q IQ2_XS | 10.68 GB | 7.08 | 11.08 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 11.26 GB | 6.70 | 10.95 |
| IQ2_M | unsloth UD-IQ2_M | 11.46 GB | 6.65 | 11.09 |
| IQ2_M | DASH-Q IQ2_M | 11.47 GB | 6.72 | 10.77 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 12.31 GB | 7.04 | 11.01 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 12.79 GB | 6.90 | 10.74 |
| Q2_K_XL | DASH-Q Q2_K_XL | 12.62 GB | 6.70 | 10.64 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Usage
llama-cli -m Qwen2.5-VL-32B-Instruct-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (Qwen/Qwen2.5-VL-32B-Instruct).