Hugging Face Model Overview
This model, DavidAU/Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Heretic-Abliterated-Heretic-Abliterated, is a 4 billion parameter Qwen3-based language model that has undergone a significant "abliteration" process using the 'Heretic' v1.0.1 method. The core objective of this modification is to drastically reduce content refusals, making it an "uncensored" model designed to answer honestly and without judgment across all use cases.
Key Capabilities & Differentiators
- Significantly Reduced Refusal Rate: Achieves a refusal rate of 5/100, a substantial improvement from the original model's 93/100, enabling unrestricted content generation.
- High Fidelity to Original Performance: Maintains a KL divergence of 0.00, ensuring the model's root/default state is preserved and its performance is not degraded by the abliteration process.
- Extended Context Window: Supports a 40960-token context length, allowing for more extensive and detailed interactions.
- Uncensored Output: Designed to generate content without refusal, including potentially sensitive or explicit material, though it may require specific directives (e.g., using slang or explicit terms) to achieve desired graphic levels.
Recommended Use Cases
This model is ideal for applications requiring an uncensored and direct AI, particularly where the goal is to generate content without typical LLM refusals. It is suitable for creative writing, roleplay, and any scenario where freedom of expression and honest, unfiltered responses are paramount. Users should be aware that for highly graphic or explicit content, some explicit direction might be needed to achieve the desired intensity.