byroneverson/gemma-2-27b-it-abliterated

byroneverson/gemma-2-27b-it-abliterated is a 27 billion parameter instruction-tuned language model, derived from Google's Gemma-2 architecture. This model is uniquely characterized by its 'abliteration' process, a CPU-only method designed to modify the model's refusal behavior. It focuses on demonstrating a novel technique for altering model responses without requiring accelerator hardware, making it suitable for exploring model safety and bias mitigation research.