PJMixers-Dev/gemma-3-1b-it-heretic-abliterated-uncensored-fixed: Uncensored Gemma 1B-IT

This model is a 1 billion parameter instruction-tuned Gemma variant, specifically modified to remove content refusals. Developed by PJMixers-Dev, it leverages the "Heretic" method (v1.0.1) to achieve a significantly reduced refusal rate of 17/100, a substantial improvement over the original Gemma 1B-IT's 99/100 refusal rate. Crucially, this uncensoring process maintains a low KL divergence of 0.09, ensuring the model's core functionality and "root state" remain largely intact, performing similarly to its original, censored counterpart.

Key Capabilities & Differentiators

• Significantly Reduced Refusal Rate: Engineered to generate content without common refusals, enabling broader creative and explicit text generation.
• High Fidelity to Original Model: A low KL divergence score indicates that the uncensoring process has not "brain-damaged" the model, preserving its original performance characteristics.
• 32K Context Length: Supports extended conversational and generative tasks.
• Targeted Uncensoring: While uncensored, the model may require explicit directives (e.g., using specific slang or terms) to generate highly graphic or explicit content at the desired intensity, as its default generation style remains "tame."

Good For

• Applications requiring unrestricted content generation, including creative writing, roleplay, or scenarios involving sensitive topics.
• Developers seeking a smaller, efficient model capable of producing diverse and uncensored outputs.
• Use cases where overcoming inherent model censorship is a primary requirement, provided users are aware of the need for explicit prompting in some instances.