Parameter‑efficient fine‑tuning that injects low‑rank matrices into pretrained weights.