Large Language Models (LLMs) have shown their power in information extraction (IE) leveraging text-to-table task. However, suffering from lost-in-the-middle problem, LLMs struggle to extract all the necessary information within longer context, leading to the missing of crucial information. Even though fine-tuned LLMs might largely alleviate this problem, it inevitably leads to more severe usage of resources, which is not affordable to all users. To address this problem, we present a novel and lightweight method, called Decomposed-Distilled Sequential Framework (DDSF), designed to enhance the capabilities of LLMs for extracting information from longer text and generating high-quality tables. Our proposed method requires lower time complexity and fewer resources without the need to fine-tune an LLM. Experiment results show that our method significantly outperforms strong baselines under four different evaluation metrics in most cases, indicating strong capability of our proposed method.