A Study on Construction of Question answering System by Fine Tuning of GPT

Abstract

Generative AI-based question-and-answer systems, like Chat GPT, are extensively utilized. These systems use general-purpose language models, prompting evaluations on domain-specific models through transfer learning. Yet, the impact of transfer learning parameters and multi-stage fine-tuning remains unclear. This study aims to uncover variations in question-and-answer systems' results caused by different transfer learning parameters and methods.