Text Summarization Project with AraGPT2

Project Overview

This project is designed to provide hands-on experience in fine-tuning the Arabic GPT-2 (AraGPT2) model for the task of text summarization. Students will work with a custom dataset and will have the opportunity to understand and implement various aspects of Natural Language Processing (NLP), particularly in the context of summarization using transformer models.

Objectives

• Understanding the fundamentals of text summarization.
• Exploring the architecture and capabilities of the AraGPT2 model.
• Fine-tuning AraGPT2 on a custom dataset for summarization tasks.
• Evaluating the performance of the fine-tuned model.

Repository Structure

• data/: Directory containing the dataset for training and validation.
• src/: Contains the source code with TODOs for students.
  • utils_data.py: Custom dataset class for handling the summarization dataset.
  • utils_tokenizer.py: Custom tokenizer for text summarization.
  • train.py: Main training script with placeholders (TODOs) for students to complete.
• main.ipynb/: Notebook for the main calls, also contains TODOs to be filled by students.
• requirements.txt: List of Python dependencies for the project.

Getting Started

You can upload the notebook and the code directly in colab =)