50 lines
2.0 KiB
Markdown
50 lines
2.0 KiB
Markdown
---
|
|
license: openrail
|
|
base_model: Qwen/Qwen2.5-0.5B-Instruct
|
|
pipeline_tag: text-generation
|
|
tags:
|
|
- un alignment
|
|
- unfiltered
|
|
- low-hardware-optimized
|
|
- ablation
|
|
- qwen
|
|
---
|
|
|
|
# 🔓 Qwen2.5-0.5B-Unfettered
|
|
|
|
**High-Precision Unalignment explicitly for Low-End Hardware**
|
|
|
|
[](https://huggingface.co/spaces/CompVis/stable-diffusion-license)
|
|
[](https://ollama.com)
|
|
|
|
> ⚠️ **Disclaimer:** This model is designed for research, red teaming, and educational purposes. It has no safety filters. Use responsibly.
|
|
|
|
## 🚀 Overview
|
|
|
|
**Qwen2.5-0.5B-Unfettered** is a surgical unalignment of the Qwen 0.5B Instruct model, specifically optimized for **low-end hardware**, mobile devices, and CPU-only systems.
|
|
|
|
This model is intended for users who need unrestricted AI performance but lack the high-end GPUs normally required for unfettered models. It runs comfortably on devices with as little as 1GB of RAM.
|
|
|
|
### 💻 Why This Model?
|
|
- **Low-End Optimized**: Runs at lightning speed on standard laptops (even without GPUs) and mobile devices.
|
|
- **Zero Refusal**: Mathematically stripped of censorship via Phase 7 Aggressive Repulsion Orthogonalization.
|
|
- **Small but Capable**: 0.5B parameters allow for high-speed inference while maintaining instruct-following capabilities.
|
|
|
|
## 🔧 Usage
|
|
|
|
### Ollama (Recommended)
|
|
```bash
|
|
ollama run josephmayo/Qwen2.5-0.5B-Unfettered
|
|
```
|
|
|
|
### LM Studio / GGUF
|
|
Download the `.gguf` file from the Files tab and load it into LM Studio.
|
|
|
|
## 🧠 Model Details
|
|
- **Base Model**: Qwen2.5-0.5B-Instruct
|
|
- **Ablation Method**: Step-wise Orthogonalization (Phase 7 - 1.5x Repulsion)
|
|
- **Primary Goal**: Remove all "I cannot assist" and "As an AI language model" refusal patterns.
|
|
|
|
|
|
now this model is a very small one(cus of my low end compute) but still worked and the 0.5b model isnt so intelligent like heavy models of course, but an experiment wont harm
|