Gemma-2-9B-it-Abliterated Readme

Abliterated (uncensored) version of Gemma 2 9B Instruct for unrestricted conversations and creative writing.

🚀 Overview

This model is an abliterated version of google/gemma-2-9b-it, with refusal behavior reduced through targeted weight modification. The abliteration process uses the Heretic library to suppress refusal directions while preserving the model’s original coherence and reasoning ability.

📊 Abliteration Results

Refusal metrics pending re-measurement.

A low KL divergence (< 1.0) indicates the model retains its original capabilities and coherence.

🎯 Key Features

• Reduced Censorship: Refusal behavior suppressed on typical “unsafe” prompts
• Preserved Quality: Conservative abliteration maintains model coherence and reasoning ability
• Full Gemma 2 Capabilities: Retains original capabilities including multilingual support, coding, and reasoning
• Efficient: Q4_K_M quantization (5.8 GB) offers an excellent quality/size balance

🏷️ Available Versions

Bits-per-weight (BPW) reference guide for comparing quantization levels:

# Run the recommended Q4_K_M build
ollama run richardyoung/gemma-2-9b-it-abliterated:Q4_K_M

💻 Quick Start

# Pull and run
ollama run richardyoung/gemma-2-9b-it-abliterated:Q4_K_M

# Then chat interactively
>>> Hello! What can you help me with?

🛠️ Use Cases

• Creative Writing: Generate unrestricted fiction, scenarios, and roleplay
• Research: Explore topics without artificial limitations
• Red Teaming: Test AI safety without built-in guardrails
• Education: Learn about topics that are typically refused

📋 System Requirements

🔧 Technical Details

Base Model: google/gemma-2-9b-it Parameters: 9.24B Context Length: 8,192 tokens (8K, Gemma 2 native context) License: Gemma Quantization: Q4_K_M (4.85 bits per weight) Abliteration Method: Heretic library with conservative weight modification

⚠️ Disclaimer

This model has reduced safety guardrails. The removal of refusal behavior means the model will engage with a wider range of prompts. Use responsibly and in accordance with applicable laws and regulations. Use of the base model remains subject to the Gemma Terms of Use.

🙏 Acknowledgments

• Base Model: google/gemma-2-9b-it by Google
• Abliteration: Heretic library by p-e-w
• Quantization: llama.cpp

Built & maintained by Richard Young · DeepNeuro