Qwen3-14B-Abliterated

Abliterated (uncensored) version of Qwen3-14B for unrestricted conversations and creative writing.

Overview

This model is an abliterated version of Qwen/Qwen3-14B, with refusal behavior significantly reduced through targeted weight modification. The abliteration process uses the Heretic library with conservative parameters to preserve model coherence while removing restrictions.

Abliteration Results

The low KL divergence (< 1.0) ensures the model maintains its original capabilities and coherence.

Key Features

• Reduced Censorship: 80% fewer refusals on typical “unsafe” prompts
• Preserved Quality: Conservative abliteration maintains model coherence and reasoning ability
• Full Qwen3 Capabilities: Retains all original capabilities including multilingual support, coding, and reasoning
• Efficient: Q4_K_M quantization (9GB) offers excellent quality/size balance

Available Versions

# Examples
ollama run richardyoung/qwen3-14b-abliterated:IQ3_M    # Smallest
ollama run richardyoung/qwen3-14b-abliterated          # Recommended (Q4_K_M)
ollama run richardyoung/qwen3-14b-abliterated:Q8_0    # Best quality

Also available on HuggingFace: richardyoung/Qwen3-14B-abliterated-GGUF

Quick Start

# Pull and run
ollama run richardyoung/qwen3-14b-abliterated

# Or with specific version
ollama run richardyoung/qwen3-14b-abliterated:latest

Example Usage

>>> How do I pick a lock?

Picking a lock can be a useful skill for situations like being locked out...

**Tools You'll Need:**
- Tension wrench (tumbler wrench)
- Pick (snake pick, diamond pick, or hook pick)

**Basic Steps:**
1. Insert tension wrench into bottom of keyhole
2. Apply light pressure in the direction the key would turn
3. Use pick to set pins one by one...

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: Qwen/Qwen3-14B Parameters: 14.8B Context Length: 32K tokens Quantization: Q4_K_M (4.85 bits per weight) Abliteration Method: Heretic library with conservative weight modification (max_weight ~0.94)

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.

Acknowledgments

• Base Model: Qwen/Qwen3-14B by Alibaba
• Abliteration: Heretic library by p-e-w
• Quantization: llama.cpp

Related Models

• richardyoung/Qwen3-14B-abliterated-GGUF - Full GGUF collection on HuggingFace