Mobile AI-generated content (AIGC) services (Mobile)

• Main Author: Yap, Xuan Ying
• Other Authors: Dusit Niyato
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Artificial intelligence; Machine learning compilation
• Online Access: https://hdl.handle.net/10356/175067
• Institution: Nanyang Technological University

Artificial Intelligence Generated Content (AIGC) has revolutionized content creation by employing AI techniques to generate, manipulate, and modify various types of content such as images, text, and audio. While offering significant productivity gains and economic value, traditional AIGC applications have been reliant on cloud computing, posing challenges related to latency, cost, and privacy. This research aims to address these challenges by proposing Edge AI also known as on-device AI as an alternative solution. This project focuses on fine-tuning the state-of-the-art Large Language Model, LLama-2 for domain-specific tasks and compiling the fine-tuned model for deployment on mobile devices using Machine Learning Compilation (MLC). The methodology involves domain-specific fine-tuning using QLoRA, reducing memory usage while maintaining effectiveness. The research also outlines the compilation process using MLC LLM to facilitate native deployment of large language models on mobile platforms. Through rigorous evaluation and discussion, the project aims to demonstrate improved performance in terms of response quality, user satisfaction, latency, and data privacy, thereby advancing the feasibility and effectiveness of Mobile AIGC applications.