基于微调GPT3.5模型的高质量小说生成研究
Investigation into generating high-quality novels via fine-tuning of GPT3.5 model
赵莉珺1
作者信息
- 1. 西藏民族大学信息工程学院,咸阳 712082
- 折叠
摘要
生成型已训练变换模型3.5(GPT3.5)特色小说文本的质量可以通过微调进一步提高,具体是通过美国开放人工智能研究中心(OpenAI)提供的GPT3.5模型的接口并对其进行改进,主要以收集高质量小说文本并对其进行信息提取、生成提示作为API数据的输入.在微调过程中的细节处理和评估方法,证明了该模型在生成小说方面的优势和局限性.相比于微调中所用到的参数调整等方法而言,提示工程对于模型的调整具有相对大的权重,因此在微调控制方法中提示设计具有重要意义.通过探索高质量小说生成模型,研究机器与小说之美的距离,这对于NLP研究是非常有意义的.
Abstract
The quality of generative trained transformation model 3.5(GPT3.5)feature novel texts can be further improved through fine-tuning,specifically through the GPT3.5 model interface provided by the Open Artificial Intelligence Research Center(OpenAI)in the United States and its improvement.The main focus is to collect high-quality novel texts,extract information from them,and generate prompts as input for API data.The detailed handling and evaluation methods during the fine-tuning process have demonstrated the advantages and limitations of this model in generating novels.Compared with methods such as parameter ad-justment used in micro adjustment,prompt engineering has relatively large weight for model adjustment,so it is of great signifi-cance to prompt design in micro adjustment control methods.It is very meaningful for NLP research to explore the generative model of high-quality novels and study the distance between machines and the beauty of novels.
关键词
微调/GPT3.5模型/小说文本生成/OpenAI/API/提示工程Key words
fine-tuning/GPT3.5 model/novel text generation/OpenAI API/prompt engineering引用本文复制引用
基金项目
西藏自治区级大学生创新训练项目(s202310695116)
出版年
2024
NETL
NSTL
万方数据