Automated alignment adjustment for LLMs — direct steering, LoRA, and MoE expert-granular abliteration, optimized via multi-objective Optuna TPE.
Fully automatic censorship removal for language models. LoRA abliteration + Optuna TPE optimization.
