license: apache-2.0
language:
- en
library_name: gguf
tags: - gemma
- moe
- gguf
- quantized
- abliterated
- multimodal
- vision
base_model: huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated
pipeline_tag: image-text-to-text
Huihui-gemma-4-26B-A4B-it-abliterated (Q4_K_M GGUF)
Converted by HSOUNA ZINOUBI
This is a Q4_K_M GGUF quantization of Huihui-gemma-4-26B-A4B-it-abliterated by huihui-ai — an uncensored version of google/gemma-4-26B-A4B-it with both thinking and non-thinking modes abliterated (first 5 layers kept intact, no ablation on the 256 experts per layer).
Details
| Property | Value |
|---|---|
| Architecture | Gemma 4 MoE |
| Total / Active | 26B / 4B |
| Quantization | Q4_K_M |
| File size | 16.8 GB |
| File | model-Q4_K_M.gguf |
Usage
llama-cli -m model-Q4_K_M.gguf -p "Your prompt" -ngl 99
For multimodal (requires mmproj):
llama-llava-cli -m model-Q4_K_M.gguf --mmproj mmproj-f16.gguf --image image.jpg -p "Describe this image" -ngl 99
Credits
- Base model: google/gemma-4-26B-A4B-it
- Abliteration: huihui-ai
- GGUF conversion: HSOUNA ZINOUBI
- llama.cpp: ggerganov/llama.cpp
Warning: This is an abliterated (uncensored) model. Safety filtering has been significantly reduced. Use responsibly.