Qwen3-4B-Thinking-2507 - DASH-Q 2-bit GGUF
llama.cpp GGUF files of Qwen/Qwen3-4B-Thinking-2507 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 |
|---|---|---|---|
Qwen3-4B-Thinking-2507-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 1.21 GB | 2.40 |
Qwen3-4B-Thinking-2507-DASHQ-IQ2_XS.gguf |
IQ2_XS | 1.40 GB | 2.78 |
Qwen3-4B-Thinking-2507-DASHQ-IQ2_M.gguf |
IQ2_M | 1.48 GB | 2.95 |
Qwen3-4B-Thinking-2507-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 1.60 GB | 3.18 |
Perplexity (lower is better)
| Type | Model | Size | WikiText-2 | C4 |
|---|---|---|---|---|
| IQ2_XXS | llama.cpp IQ2_XXS (imatrix) | 1.25 GB | 22.42 | 34.84 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 1.26 GB | 25.43 | 33.53 |
| IQ2_XXS | DASH-Q IQ2_XXS | 1.21 GB | 14.18 | 24.57 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 1.35 GB | 15.61 | 24.69 |
| IQ2_XS | DASH-Q IQ2_XS | 1.40 GB | 12.01 | 20.33 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 1.51 GB | 12.91 | 21.15 |
| IQ2_M | unsloth UD-IQ2_M | 1.53 GB | 14.34 | 21.64 |
| IQ2_M | DASH-Q IQ2_M | 1.48 GB | 11.66 | 19.32 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 1.67 GB | 12.40 | 20.24 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 1.70 GB | 12.03 | 19.45 |
| Q2_K_XL | DASH-Q Q2_K_XL | 1.60 GB | 11.18 | 18.49 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Usage
llama-cli -m Qwen3-4B-Thinking-2507-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (Qwen/Qwen3-4B-Thinking-2507).