gemma-4-26B-A4B-it - DASH-Q 2-bit GGUF
llama.cpp GGUF files of google/gemma-4-26B-A4B-it 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 |
|---|---|---|---|
gemma-4-26B-A4B-it-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 10.10 GB | 3.20 |
gemma-4-26B-A4B-it-DASHQ-IQ2_XS.gguf |
IQ2_XS | 10.60 GB | 3.36 |
gemma-4-26B-A4B-it-DASHQ-IQ2_M.gguf |
IQ2_M | 11.11 GB | 3.52 |
gemma-4-26B-A4B-it-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 11.27 GB | 3.57 |
Text only: the vision tower is not included.
Quality vs. F16 (KL divergence)
Raw-text perplexity is not a meaningful metric for gemma-4-it (the F16 model itself scores in the thousands on WikiText-2 while generating normally),
so quality is reported as KL divergence of the next-token distribution to F16 (llama-perplexity --kl-divergence, WikiText-2, 4 x 2048 tokens):
| Type | Model | Size | Mean KLD | Median KLD | Top-1 token agreement |
|---|---|---|---|---|---|
| IQ2_XXS | llama.cpp IQ2_XXS (imatrix) | 9.28 GB | 6.34 | 5.29 | 24.1% |
| IQ2_XXS | unsloth UD-IQ2_XXS | 9.92 GB | 2.27 | 1.02 | 50.8% |
| IQ2_XXS | DASH-Q IQ2_XXS | 10.10 GB | 2.25 | 1.01 | 51.5% |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 9.80 GB | 4.91 | 3.92 | 30.4% |
| IQ2_XS | DASH-Q IQ2_XS | 10.60 GB | 2.16 | 0.87 | 52.1% |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 10.36 GB | 3.79 | 2.96 | 33.3% |
| IQ2_M | unsloth UD-IQ2_M | 10.01 GB | 2.28 | 1.02 | 51.1% |
| IQ2_M | DASH-Q IQ2_M | 11.11 GB | 1.85 | 0.72 | 55.2% |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 10.58 GB | 5.25 | 4.48 | 28.0% |
| Q2_K_XL | DASH-Q (this file) | 11.27 GB | 1.83 | 0.75 | 54.6% |
Usage
llama-cli -m gemma-4-26B-A4B-it-DASHQ-IQ2_M.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (google/gemma-4-26B-A4B-it).