---
pipeline_tag: image-text-to-text
base_model: huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated
base_model_relation: quantized
library_name: llama.cpp
tags:
  - gguf
  - mxfp4
  - quantized
  - multimodal
  - abliterated
  - uncensored
---

These are **UD quantizations** of [huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated](https://huggingface.co/huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated), packaged for **llama.cpp / GGUF** inference.

## Quick Start
1. Download the latest release of **llama.cpp**.
2. Download your preferred model variant from the files below.
3. Use the corresponding `mmproj` file for multimodal inference.

## Which version should I choose?
These variants are built using a **UD tensor distribution / quantization recipe** to preserve as much quality as possible while reducing memory usage.

Choose based on your hardware and quality target:

| Variant | Quality | Speed | VRAM / RAM | Recommendation |
| :-- | :-- | :-- | :-- | :-- |
| **BF16** | Highest | Slowest / variable | Highest | Best quality, if your hardware can handle it |
| **F16** | High | Faster | High | Good balance of quality and speed |
| **Q8** | Solid | Fastest | Lowest of these three | Best choice for efficiency |

## Notes
- This repo is a **quantized release**, not a finetune.
- Base model: [huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated](https://huggingface.co/huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated)
- Runtime: **llama.cpp**