Efficient Pruning of Large Language Model with Adaptive Estimation Fusion