Uncensored Qwen3-Based Model for Unrestricted Content Generation

DavidAU/Qwen3-4B-Thinking-2507-Gemini-2.5-Flash-Lite-Preview-Distill-Heretic-Abliterated is a Qwen3-based model that has undergone a "de-censoring" process using the Heretic v1.0.1 method. This process aims to remove content refusals while preserving the model's original quality.

Key Characteristics

• Abliterated/Uncensored: Achieves a significantly low refusal rate of 1/100, down from an original 53/100, ensuring the model answers honestly and without judgment.
• High Fidelity: Maintains a KL divergence of 0.00, indicating that the de-censoring process has not damaged the model's root state or performance.
• Extended Context: Features a 256k context length, allowing for processing and generation of longer texts.
• Flexible Content Generation: Designed to generate content across all use cases, including potentially sensitive or explicit topics, without inherent refusals.

Usage Considerations

While uncensored, the model may require explicit direction (e.g., specifying slang or graphic terms) to generate content at desired levels of intensity for certain topics. This model is part of the broader "Qwen3-24B-A4B-Freedom-Thinking-Abliterated-Heretic-NEO" initiative, emphasizing user freedom and control. Users can adjust settings like "Smoothing_factor" (1.5 recommended) in interfaces like KoboldCpp or oobabooga for smoother operation, especially for chat and roleplay scenarios.