Distillation Matters: Empowering Sequential Recommenders to Match the Performance of Large Language Model