Parameter‑efficient fine tuning approach that combines low‑rank adaptation with quantized base weights to cut VRAM use.