This paper proposes a novel inner-loop dataset distillation framework , which automatically adjusts the truncation position and window size, and utilizes a low-rank Hessian approximation to reduce computational overhead.