LBLLM: Lightweight Binarization of Large Language Models via Three-Stage Distillation | ScienceToStartup