Model Overview
DavidAU/Qwen3-4B-Apollo-V0.1-4B-Thinking-Heretic-Abliterated is a Qwen3-4B model that has undergone a "de-censoring" process using the Heretic v1.0.1 method. The primary goal of this modification is to drastically reduce the model's refusal rate, which has been brought down from an original 96/100 to 29/100. This process also prioritizes maintaining the model's original performance and integrity, as indicated by a low KL divergence of 0.09, ensuring it is not "brain damaged."
Key Capabilities
- Uncensored Content Generation: Designed to provide honest and unrestricted answers across all topics, without refusal or judgment.
- High Context Length: Features a 256k context window, allowing for extensive and detailed interactions.
- Integrity Preserved: The Heretic method ensures that while censorship is removed, the model's core functionality and quality are maintained.
- Flexible Content Creation: Capable of generating various content types, including those typically restricted, though it may require explicit directives (e.g., using specific slang or terms) to achieve desired graphic or explicit levels.
Good For
- Unrestricted Use Cases: Ideal for applications requiring complete freedom in content generation, including creative writing, roleplay, and scenarios where typical LLM censorship is undesirable.
- Exploration of Sensitive Topics: Suitable for research or applications that need to explore or generate content on sensitive or controversial subjects without built-in refusals.
- Customizable Output: Users can "push" the model with specific instructions to achieve desired levels of detail, graphic content, or language style.
For optimal performance, users are advised to adjust settings like "Smoothing_factor" in interfaces like KoboldCpp or oobabooga/text-generation-webui, and to consult the provided guides for advanced parameters and samplers.