Development of a Music Education Framework Using Large Language Models (LLMs)

Abstract

This thesis explores the effectiveness of Large Language Models (LLMs) in enhancing educational methodologies, particularly focusing on personalized learning experiences in music education. Initially, a comprehensive literature review was conducted to establish the theoretical foundation and identify gaps in the current application of Large Language Models in education. Subsequently, employing a quantitative approach, the study utilized the Supervised Fine-Tuning QLoRA approach to adapt the Llama2-chat model to respond accurately to music educational queries. The Results showed the fine-tuned model with the instruction dataset provides some good results on the provided prompts. The performance of the model was evaluated using standard metrics such as BERTScore, F1 Score, and Exact_Match, which confirmed the model’s efficacy in providing accurate and contextually appropriate responses. While the findings confirm the potential of integrating LLMs into educational frameworks, they also highlight some limitations, such as the need for continuous model training to adapt to evolving and diverse musical content and creativity. This study establishes a basis for future research, suggesting the exploration of symbolic music understanding models like MusicBERT and LLMs integration within music education.