deewu0809/Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Heretic-Abliterated-Heretic-Abliterated
The deewu0809/Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Heretic-Abliterated-Heretic-Abliterated is a 4 billion parameter Qwen3-based language model, fine-tuned using the Heretic method to significantly reduce refusal rates while maintaining model integrity. It features a 32768 token context length and is specifically designed for uncensored, honest responses across all use cases, aiming for maximal freedom in content generation.
Loading preview...
Model Overview
This model, based on the Qwen3 architecture with 4 billion parameters, has been processed using the "Heretic" method (v1.0.1) to achieve a significantly reduced refusal rate of 5/100, down from an original 93/100. The Heretic method ensures that the model's root state remains largely undamaged, as indicated by a KL divergence of 0.00, meaning it performs as well as the original model while being "abliterated" (de-censored).
Key Capabilities
- Uncensored Content Generation: Designed to provide honest and direct answers without refusals, suitable for a wide range of topics including those typically censored.
- High Context Length: Supports a substantial context window of 32768 tokens, allowing for detailed and extended interactions.
- Integrity Preserved: The abliteration process maintains the model's original performance and quality, as validated by a near-zero KL divergence.
Usage Considerations
While uncensored, the model may require explicit directives (e.g., including specific slang or terms) to generate content at expected graphic or explicit levels, especially for x-rated or highly descriptive content. It will not refuse requests but might produce "bland" output without sufficient prompting. Users can adjust settings like "Smoothing_factor" (1.5) in interfaces like KoboldCpp or oobabooga/text-generation-webui for smoother operation and improved chat/roleplay experiences.