Abstract: This paper focused on the problems of poor network attack category detection performance and low network security defense performance in existing technologies, and conducted research on network security defense technology and applications based on Large Language Model (LLM) and Transformer mechanism. It introduced LLM and Transformer mechanisms to construct a network security defense model, used LLM to collect and preprocess the training data of the constructed model, established a mapping relationship between training data and network attack categories based on the Transformer mechanism, and completed the pre-training of the model construction. Based on the loss function, the paper fined tune the parameters of the constructed model, and based on the pre-training results of the constructed model and the fine tuning results of the parameters, the paper determined the network attack category to implement effective detection and defense of the network attack category. The experimental results show that the network attack category detection effect of this technology is superior to the comparative technology, and the maximum F1 score can reach 0.9.