Overview

MEDFIT-LLM-3B is a specialized 3.2 billion parameter language model, fine-tuned from Meta's Llama-3.2-3B-Instruct. Developed by Aditya Karnam Gururaj Rao, Arjun Jaggi, and Sonam Naidu, this model is specifically optimized for healthcare and medical question-answering (QA) applications. It leverages LoRA (Low-Rank Adaptation) techniques on a curated dataset of 6,444 healthcare-related question-answer pairs, enhancing its ability to provide direct and structured medical information.

Key Capabilities & Performance

• Enhanced Direct Answers: Achieves a 30 percentage point increase in direct answer rate (from 6.0% to 36.0%) compared to the base model.
• Improved Response Structure: Shows an 18% increase in the use of numbered lists, leading to better organized and more comprehensive answers.
• Domain Specialization: Optimized for medical information dissemination and patient education within healthcare chatbot contexts.
• Efficient Fine-tuning: Utilizes LoRA with the MLX framework, minimizing computational requirements while maximizing domain-specific performance.

Ideal Use Cases

MEDFIT-LLM-3B is designed for applications requiring accurate and structured medical information delivery. It is particularly well-suited for:

• Medical Question Answering: Providing precise responses to healthcare queries.
• Patient Education: Delivering easy-to-understand medical information.
• Healthcare Chatbot Applications: Serving as the core for conversational agents in healthcare.
• Integration into: Healthcare mobile apps, medical information systems, and telemedicine platforms.

Important Limitations

It is crucial to note that MEDFIT-LLM-3B is not intended for medical diagnosis, treatment recommendations, or emergency situations. It should always be used as a supplementary tool with human oversight and never as a substitute for professional medical advice.