Automated alignment adjustment for LLMs — direct steering, LoRA, and MoE expert-granular abliteration, optimized via multi-objective Optuna TPE.
An Easy-to-use Steering Framework for Editing Large Language Models
