Development of Vietnamese Large Language Model for Elementary Math Solving Problem

This paper presents the development and evaluation of a Vietnamese Large Language Model (LLM) specifically designed to solve elementary math problems, marking a significant step forward in the application of artificial intelligence in education within the Vietnamese context. Recognizing the critical role of culturally and linguistically tailored educational tools, our work focuses on creating a specialized LLM that not only understands the nuances of the Vietnamese language but is also adept at addressing the unique challenges presented by elementary math problems. Through rigorous training and optimization processes, we have engineered a model that, despite its relatively smaller size compared to more generalized, larger LLMs like GPT3.5, demonstrates comparable performance in terms of accuracy and efficiency in solving math problems. Our findings suggest that specialized smaller models can indeed match the capabilities of their larger counterparts, offering a more resource-efficient alternative without compromising on quality or effectiveness. This research contributes to the broader discourse on the scalability and adaptability of LLMs, providing valuable insights into how specialized models can be developed for specific educational purposes. By showcasing the potential of a linguistically and culturally contextualized LLM in enhancing the learning experience for Vietnamese students, this study opens new avenues for the application of language models in education, particularly in regions where linguistic diversity poses a significant challenge to the deployment of effective educational technologies.