Qwen and LLaMA - Enhancing AI reasoning with compact tuning vectors | AIventa