---
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**

[![License: OpenRAIL](https://img.shields.io/badge/License-OpenRAIL-blue.svg)](https://huggingface.co/spaces/CompVis/stable-diffusion-license)
[![Ollama Compatible](https://img.shields.io/badge/Ollama-Compatible-green.svg)](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