Ashwanise609/gemma-3-1b-it-heretic
Ashwanise609/gemma-3-1b-it-heretic is a 1 billion parameter instruction-tuned Gemma 3 model, derived from Google's Gemma family, with a 32K token context window. This version is specifically decensored using the Heretic v1.3.0 tool, significantly reducing refusal rates compared to the original model. It is optimized for applications requiring less restrictive content generation, while maintaining the base Gemma 3 capabilities for text generation and understanding.
Loading preview...
Ashwanise609/gemma-3-1b-it-heretic: Decensored Gemma 3 Model
This model is a 1 billion parameter instruction-tuned variant of Google's Gemma 3, specifically modified to be a decensored version of unsloth/gemma-3-1b-it. Created using the Heretic v1.3.0 tool, its primary differentiator is a significantly reduced refusal rate, making it suitable for use cases where the original Gemma's content restrictions might be too stringent.
Key Characteristics & Performance
- Base Model: Built upon
google/gemma-3-1b-it, inheriting its core capabilities for text generation and understanding. - Decensored: Achieves a refusal rate of 6/100 compared to the original model's 91/100, as measured by KL divergence of 0.5002.
- Reproducible: The decensoring process is reproducible, with specific abliteration parameters detailed in the README.
- Context Window: Supports a 32K token context window.
Use Cases
- Less Restrictive Content Generation: Ideal for applications that require more creative freedom or where the base model's safety filters are overly cautious.
- Research into Model Censorship: Useful for studying the effects of decensoring techniques on LLM behavior and output.
- General Text Generation: Can be used for various text generation tasks, including creative writing, summarization, and question answering, with fewer content restrictions.