A Web Application for a Cost-Effective Fine-Tuning of Open-Source LLMs in Education

Many Generative Artificial Intelligence (GenAI) applications built on Large Language Models (LLMs) emerged in 2023 causing a great impact in the educational landscape. However, most of these GenAI applications require a subscription to access the most advanced models and functionalities. Therefore, the search for cost-effective solutions becomes an important concern, especially for instructors and educational institutions with limited resources. This article introduces a Web Application aimed at facilitating instructors in fine-tuning open-source LLMs and subsequently posing questions to them. Instructors only need to upload a dataset into the Web Application to fine-tune the opensource LLM, specifically Llama 2. This web application was developed using open-source tools (Hugging Face and Langchain), and techniques to reduce hardware resource consumption (LoRA and QLoRA). Preliminary results from the experiments conducted show that LLMs provide more accurate responses when fine-tuned for a specific task through the Web Application. These are the first steps in providing cost-effective GenAI solutions for instructors and educational institutions using opensource tools.