Pipeline for uncensoring gpt-oss-20b with external uncensored teachers (LLaMA-2/3). Includes refusal/jailbreak prompt libraries, automated response generation, cleaning, dataset building, and LoRA/QLoRA fine-tuning. Produces streamlined uncensored checkpoints with evaluation scripts, configs, and adapter management for efficient local deployment.