Fine-Tuning Large Language Models for the Electric Power Industry with Specialized Chinese Electric Power Corpora
Jiajia Han, Tao Yang, Cai Zhang, Han Jiang · 2024
The advent of large language models has exerted a profound impact across various sectors. Despite significant strides in general applications and certain domain-specific areas, existing general language models exhibit limitations in addressing the specific requirements of the electric power industry. Especially when processing Chinese content, their performance often falls short compared to handling English content. In response to this challenge, our research team collected and organized a large volume of Chinese literature related to the electric power domain and established a specialized corpus. Leveraging this corpus, we combined manually edited question-answer pairs with the generative capabilities of ChatGPT to construct an instruction dataset for the electric power domain. Based on this dataset, we successfully trained a large language model focusing on the electric power industry. In addition, we developed an evaluation dataset to screen out large language model architectures that perform well in knowledge storage and presentation capabilities. After comprehensive evaluation, our large language model in electric power domain shows superior performance in question answering tasks within the professional field.