Fine-Tuned Language Models in Healthcare: Empowering Affordable Medical Consultations

Abstract

The increasing demand for accessible healthcare, particularly in regions like India with high costs and limited resources, motivates the exploration of efficient AI-powered solutions. Our work explores the use of Large Language Models (LLM) fine-tuned on a self-curated dataset consisting of patientdoctor conversations for preliminary medical consultancy, by utilizing approaches such as Low-Rank Adaptation (LoRA) in conjunction with compression methods like 4-bit quantization to minimize the model's computational footprint for deployment on mobile devices. Our model achieves an average accuracy of 59.05% across a suite of medical benchmarks, outperforming comparable 3 billion parameter models and even some larger 7 billion parameter models while operating at a significantly lower precision. This paper highlights the effectiveness of combining quantization-aware fine-tuning towards creating accurate and efficient LLMs for specialized domains, even in resourceconstrained environments, which contributes to the development of accessible and affordable AI-powered healthcare solutions.

Ähnliche Arbeiten