Abstract:With the rapid development of artificial intelligence technology, large language model have a significant impact on various industries, Transformer architecture large model have received widespread attention due to their unique advantages and potentials, but there are also many questions and doubts. This paper firstly introduces the development lineage of large model, and the status quo and problems of Transformer architecture large models, and then focuses on the technical architecture and experimental result of the non-Transformer architecture Yan architecture large model, especially the differences and advantages between them and the Transformer architecture models in the aspects of training effect, throughput, and computational resource consumption. In addition, the application construction of Yan architecture large model in the field of intelligent customer service of State Grid material suppliers was discussed，and how to optimize the algorithm architecture to provide faster response speed, more accurate semantic comprehension ability and optimize the customer service experience. Finally, it looks forward to the application prospect and development trend of non-Transformer architecture big models in the future intelligent customer service and other natural language processing fields, pointing out its important role in promoting the progress of AI technology and service innovation.