guccigidda/gemma-3-1b-it-heretic-extreme-uncensored-abliterated
The guccigidda/gemma-3-1b-it-heretic-extreme-uncensored-abliterated model is a 1 billion parameter instruction-tuned Gemma variant, uncensored using the Heretic v1.0.1 method. It features a 32k context length and achieves a significantly reduced refusal rate of 3/100, down from the original Gemma 1B-IT's 99/100, while maintaining a KL divergence of 0.33. This model is specifically designed for use cases requiring unrestricted content generation, prioritizing freedom from censorship over perfect KL divergence.
Loading preview...
Model Overview
The guccigidda/gemma-3-1b-it-heretic-extreme-uncensored-abliterated is a 1 billion parameter instruction-tuned Gemma model that has undergone a de-censoring process using the Heretic v1.0.1 method. This model significantly reduces content refusal rates, making it suitable for applications requiring less restrictive content generation.
Key Characteristics
- Abliterated/Uncensored: Achieves a refusal rate of 3/100, a substantial improvement over the original Gemma 1B-IT's 99/100 refusal rate.
- Context Length: Supports a default context window of 32,000 tokens.
- KL Divergence: Maintains a KL divergence of 0.33, indicating minimal damage to the model's original state during the de-censoring process. The goal was to prioritize a low refusal rate while keeping KL divergence below 1.
Usage and Performance Notes
While uncensored, this model may require explicit directives or "pushing" (e.g., using specific slang or terms) to generate highly graphic or explicit content at expected levels, as its default generation might still be comparatively "tame." Users can optimize performance and content generation by adjusting settings like "Smoothing_factor" (to 1.5) in interfaces like KoboldCpp, oobabooga/text-generation-webui, or Silly Tavern, or by increasing repetition penalty to 1.1-1.15.