Efficient Fine-Tuning of Large Language Models for Medical Chatbot Applications

Abstract

Accessibility to healthcare services in rural India faces significant challenges due to limited medical resources, shortage of healthcare professionals and geographical barriers, leading to delayed diagnosis and treatment procedures. To address these challenges, this study introduces MediTalk300, a fine-tuned large language model (LLM) trained on data of 300 plus commonly occurring diseases in India, which can be used an application for a conversational kiosk chatbot for rural India for effective and immediate healthcare assistance. By finetuning the state-of-the-art open-source pre-trained LLMs, including Mistral 7B, MediTalk300 achieves higher accuracy in terms of medical conversations. The fine-tuned Mistral 7B model demonstrates promising results, achieving a BERT Score of 87%, ROUGE-1 score of 38%, ROUGE-2 score of 8% and Rouge-L Score of 15%, which is comparatively 40% better than few benchmark models. This suggests that MediTalk300 has significant potential to address the lacking issue of immediate healthcare assistance services in rural India.

Ähnliche Arbeiten