MedicalQA-T5 Model

Project Overview

This project involves fine-tuning the T5 transformer model for medical question-answering tasks. The model is trained on a domain-specific dataset, enabling it to generate accurate and contextually relevant medical responses.

Installation

1. Clone the repository:

   git clone https://github.com/abdulvahapmutlu/medicalqa-t5-model.git
   

2. Install dependencies:

   pip install -r requirements.txt
   

Usage

Data Preprocessing

Run the preprocessing.py script to prepare the data and train the SentencePiece model.

python scripts/preprocessing.py

Model Training

Train the T5 model on the preprocessed dataset:

python scripts/training.py

Text Generation

Generate medical answers using the fine-tuned model:

python scripts/generation.py

Model Performance

• Final Training Loss: ~0.96
• Final Validation Loss: ~0.85

The model consistently improved across epochs, demonstrating effective learning and generalization.

Future Work

• Expand the dataset for broader medical domains.
• Experiment with larger T5 models or alternative transformer architectures.
• Deploy the model in a web-based application for real-time medical Q&A.

License

This project is licensed under the MIT License.