初始化项目，由ModelHub XC社区提供模型

Model: josephmayo/Qwen2.5-0.5B-Unfettered
Source: Original Platform

README.md

@@ -0,0 +1,49 @@
+---
+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