huihui-ai/gemma-3-12b-it-abliterated

Overview

This model, huihui-ai/gemma-3-12b-it-abliterated, is a modified version of Google's gemma-3-12b-it instruction-tuned model. Its primary distinction is the removal of refusal behaviors, achieved through a process called "abliteration" (detailed in the remove-refusals-with-transformers project). This modification aims to prevent the model from generating responses like "I'm programmed to be a safe and helpful AI assistant. I cannot fulfill your request to..."

Key Capabilities

• Uncensored Responses: Designed to provide direct answers without safety-aligned refusals.
• Multimodal Support: Inherits the multimodal capabilities of the base Gemma-3 model, processing both text and image inputs.
• Instruction-Following: Retains the instruction-tuned nature of the original Gemma-3-12B-IT.

Usage and Integration

• The model can be loaded and used with Hugging Face's transformers library, supporting Gemma3ForConditionalGeneration.
• It is also available for use with Ollama, with a specific huihui_ai/gemma3-abliterated:12b version provided.
• The abliteration process focused solely on the text component, not the image processing part.

Good For

• Use cases where unfiltered or direct AI responses are preferred or required.
• Developers experimenting with model safety and refusal mechanisms.
• Applications needing a multimodal model that avoids common AI safety disclaimers.