Optimization of Chinese Diplomatic Rhetoric Generation Based on LoRA Efficient Fine-Tuning

Diplomatic rhetoric is a wording method used to express the position of national government in diplomatic occasions, and it is a language tool to reflect a country's foreign policy and defend a country's national interests. There are problems such as lack of rigor and standardized terminology in the text generation of Chinese diplomatic rhetoric by large language models. In this paper, we focus on the generation scenario of Chinese diplomatic rhetoric, and propose an efficient fine-tuning optimization scheme of parameters based on LoRA. Firstly, LoRA parameters are efficiently fine-tuned for large language models by utilizing high-quality samples from the Alpaca-format Chinese diplomatic rhetoric dataset, and the experimental method of multi-model fine-tuning performance comparison is adopted to verify the significant advantages of LoRA technology in text generation in the professional domain. Secondly, LoRA technology is innovatively applied to diplomatic language generation scenarios to achieve efficient parameter updating through low-rank matrix decomposition technology, which optimizes the generation quality while preserving the integrity of pre-trained knowledge. The experimental results show that the LoRA efficient fine-tuning technique can significantly improve the large language model's grasp of Chinese context and diplomatic language characteristics.