Evaluating the Generation Capabilities of Large Chinese Language Models