grimjim/gemma-3-12b-it-biprojected-abliterated

grimjim/gemma-3-12b-it-biprojected-abliterated is a model derived from Google's Gemma-3-12b-it, modified using "projected abliteration" to significantly reduce refusal rates while maintaining safety awareness. This 12 billion parameter model is designed for applications requiring less frequent content refusal responses. It is suitable for use cases where a more permissive response style is desired without completely sacrificing safety considerations.