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

grimjim/gemma-3-12b-it-norm-preserved-biprojected-abliterated is a 12 billion parameter instruction-tuned model derived from Google's Gemma-3-12b-it, featuring a 32768 token context length. It utilizes 'norm-preserved biprojected abliteration' to significantly reduce refusal rates while maintaining safety awareness. This model is optimized for scenarios requiring reduced content refusal without subsequent fine-tuning to repair potential damage.