Monisha27/medical-llm-sft-demo

🩺 Medical LLM – Fine-Tuned DialoGPT

This project fine-tunes microsoft/DialoGPT-medium on a medical Q&A dataset using Unsloth + LoRA for efficient training. The goal is to make the model better at answering health-related questions compared to the base model.

🚀 Project Overview

Base Model: microsoft/DialoGPT-medium

Fine-tuning Framework: Unsloth

Method: Parameter-Efficient Fine-Tuning (LoRA, 16-bit)

Dataset: Custom Medical Q&A (500 examples used for demo)

Deployment: Hugging Face Model + Space Demo

⚙️ Training Setup

Hardware: Tesla T4 GPU (Colab)

Max Seq Length: 2048

Batch Size (effective): 8 (2 × 4 grad accumulation)

Epochs: 1 (demo)

Trainable Parameters: 1.7% of total (6.29M)

📊 Results Example 1

Question: What are the symptoms of pneumonia?

Base Model: What are the symptoms of pneumonia?

Fine-Tuned Model: Cough, fever, chills, and difficulty breathing are common symptoms of pneumonia. ✅

Example 2

Question: How is diabetes diagnosed?

Base Model: bloodwork and a banana. ❌

Fine-Tuned Model: Diabetes is usually diagnosed with blood tests like fasting blood sugar and HbA1c. ✅

💻 Usage Load from Hugging Face from transformers import pipeline

pipe = pipeline( "text-generation", model="Monisha27/medical-llm-sft-demo" )

print(pipe("What are the symptoms of pneumonia?", max_new_tokens=50))

📂 Repository Structure

config.json – model configuration

tokenizer.json – tokenizer

pytorch_model.bin – model weights (LoRA merged)

README.md – this documentation

🌐 Demo

Try the model live on Hugging Face Spaces: 👉 Medical LLM Chatbot

📌 Notes

This model is not a replacement for professional medical advice.

Outputs may contain mistakes. Always consult a doctor for real medical concerns.

✨ Developed by Monisha Bharadwaj