Ophtimus-LLM: Development of a Specialized Large Language Model for Ophthalmology

Seung Ju Baek, Kuk Jin Jang, Sooyong Jang, Hyonyoung Choi, Minwook Kwon, Yong Seop Han, Seongjin Lee, Jin Hyun Kim, Insup Lee

Keywords: Generative AI, Large language Models, Ophthalmology, Health AI, Domain-specific AI

Abstract: The development of domain-specific language models has become increasingly important in healthcare, where the complexity and precision of medical knowledge often exceed the capabilities of general-purpose large language models (LLMs). This study introduces Ophtimus-LLM, a compact 8 billion-parameter LLM tailored for ophthalmology. Key findings demonstrate that scalability laws observed in larger models also hold for our smaller, domain-specific LLMs, suggesting that well-designed compact models can achieve high performance. Additionally, the study highlights the critical role of data quality in boosting model accuracy, with significant gains observed when training on domain-relevant content. Ophtimus-LLM exemplifies the potential of specialized LLMs to provide efficient, accessible, and high-performing tools for advancing medical AI while addressing challenges of scalability and equity in healthcare technology.

Submission Number: 72