Modelpublic

dungnvt/Huihui-Qwopus3.5-9B-v3-abliterated-GGUF

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes193downloads

Model Card

basemodel: huihui-ai/Huihui-Qwopus3.5-9B-v3-abliterated libraryname: gguf license: apache-2.0 tags:

—qwen
—qwen3.5
—gguf
—unsloth
—abliterated
—uncensored
—vision pipeline_tag: image-text-to-text ---

Huihui-Qwopus3.5-9B-v3-abliterated - GGUF (Q80/Q6K)

This repository contains high-quality GGUF quantizations of the Huihui-Qwopus3.5-9B-v3-abliterated model.

🚀 Usage with Ollama

You can run this model locally with full vision support using Ollama.

1. Download the files Keep both .gguf files and the Modelfile in the same directory.

2. Create the model Open your terminal in that folder and run:

bash

ollama create qwopus9b-vision -f Modelfile

3. Run and Analyze

bash

ollama run qwopus9b-vision

Note: To analyze an image, simply provide the local path to the image file in your prompt (e.g., `/path/to/screenshot.png Describe the errors in this image`).

💡 Technical Recommendations

—Minimum VRAM: 12 GB.
—Recommended VRAM: 16 GB (especially for contexts above 32k tokens).
—KV Cache Optimization: To handle extremely long contexts (>64k tokens) on 16GB cards, consider setting the environment variable OLLAMA_KV_CACHE_TYPE=q8_0.

🛡️ About the "Abliterated" Version

This model has undergone an "abliteration" process to remove built-in refusal mechanisms. It will follow instructions more directly and is less likely to refuse requests due to safety filters, making it highly effective for specialized tasks like security auditing or uncensored creative writing.

🙏 Credits

Special thanks to huihui-ai for the original fine-tuning and abliteration, and the Unsloth team for their state-of-the-art quantization tools.

---
!!!From ISeLab with Love!!!