pegasus-dialogue-summarization

Text Summarization Project 🚀

Fine-tuning PEGASUS on the SAMSum dataset to summarize conversational dialogues into short, meaningful text summaries.
Built with a complete ML pipeline: Research ➔ Modular Codebase ➔ API Deployment ➔ Dockerization.

---

📚 Problem Statement

Modern conversations (chat, emails, messages) often contain redundant and lengthy information.
Text summarization helps extract key information efficiently, saving time and effort.

This project focuses on building a dialogue summarization system using Transformer-based models.

---

🎯 Project Goals

• Fine-tune google/pegasus model on real-world chat conversations (SAMSum dataset).
• Build an end-to-end machine learning pipeline: from data ingestion to model evaluation.
• Deploy the trained summarizer via an API.
• Containerize the entire application using Docker for easy deployment.

---

🏗️ Project Architecture

Stage	Details
Research Notebooks	Data ingestion, validation, transformation, training, evaluation (modular Jupyter notebooks).
Training Pipeline (Python scripts)	Modularized under src/textSummarizer/ using Clean Code principles.
Configuration Management	params.yaml (training parameters) and config/config.yaml (paths/settings).
Model Serving	app.py to deploy model API (FastAPI based).
Docker Containerization	Dockerfile for building and running the app easily.

---

📂 Project Structure

Text_Summarization_Project/
├── .github/workflows/        # CI/CD workflows (future-ready)
├── config/                   # Configuration files
│    └── config.yaml
├── research/                 # Jupyter notebooks for research and experimentation
│    ├── 01_data_ingestion.ipynb
│    ├── 02_data_validation.ipynb
│    ├── 03_data_transformation.ipynb
│    ├── 04_model_trainer.ipynb
│    ├── 05_model_evaluation.ipynb
│    ├── Text_Summarization.ipynb
│    └── trials.ipynb
├── src/textSummarizer/        # Core source code
│    ├── components/           # Modular components (ingestion, training etc.)
│    ├── config/               # Configuration parsers
│    ├── constants/            # Constant values
│    ├── entity/               # Data schemas
│    ├── logging/              # Logging utilities
│    ├── pipeline/             # Training and prediction pipelines
│    ├── utils/                # Helper functions
│    └── __init__.py
├── app.py                     # Serve model API
├── main.py                    # Main runner
├── Dockerfile                 # Docker setup
├── setup.py                   # Package installation
├── requirements.txt           # Project dependencies
├── params.yaml                # Hyperparameters for training
├── README.md                  # Project documentation
└── template.py                # Folder structure generator

---

🔥 Key Features

• Fine-tuning PEGASUS-large model using Hugging Face Transformers.
• Research-first Approach with modular notebooks.
• Production-Ready Codebase inside src/textSummarizer/.
• Config-Driven Development using yaml files.
• Containerization with Docker for easy deployment.
• Scalable Architecture – ready for extending to larger datasets or newer models.

---

🛠 Tech Stack

• Python 3.8+
• PyTorch
• Hugging Face Transformers
• Hugging Face Datasets
• Flask (for API serving)
• Docker
• YAML Configuration Management
• GitHub Actions (for future CI/CD)

---

🚀 How to Run Locally

1. Clone the repository

git clone https://github.com/ShalinVachheta017/Text_Summarization_Project.git
cd Text_Summarization_Project

2. Install Dependencies

pip install -r requirements.txt

3. Train the Model

python main.py

4. Run the API Server

python app.py

The server will start at http://127.0.0.1:5000/.

---

Sample Input:

Alex: Are we still on for 6 PM?
Jordan: Running 10 minutes late, but yes!

Generated Summary:

Alex and Jordan plan to meet at 6 PM, with a slight delay.

---

⚡ Future Work

• Deploy model on Hugging Face Spaces / Streamlit.
• Fine-tune PEGASUS on larger dialogue datasets (e.g., Reddit conversations).
• Experiment with model distillation to reduce size and improve speed.
• Add MLflow tracking for experiments and metrics.
• Add CI/CD pipeline for automatic deployment.

---

🙋‍♂️ Author

• Shalin Vachheta
• GitHub | LinkedIn

---

"Summarizing conversations is not just compression, it's distilling meaning." 🌟

---