A Simple and Effective Pruning Approach for Large Language Models