T5 Text Summarizer 🚀

A GPU-accelerated text summarization web application powered by a fine-tuned T5 (Text-to-Text Transfer Transformer) model, served via a FastAPI backend with a premium, dark-themed UI.

Features · Architecture · Quick Start · API Reference · Project Structure

---

✨ Features

• 🤖 Fine-tuned T5 Model — Custom-trained T5-base model (t5-base) optimized for abstractive text summarization
• ⚡ GPU Acceleration — Enforced CUDA inference via PyTorch for fast, real-time summaries
• 🌐 FastAPI Backend — High-performance async REST API with automatic OpenAPI docs
• 🎨 Premium Web UI — Dark-themed, glassmorphism interface with Three.js animated shader background
• 📏 Adjustable Length — Real-time slider to control max_token_length (50–500 tokens)
• 📋 One-click Copy — Instantly copy the generated summary to clipboard
• 🧹 Text Preprocessing — Removes HTML tags, extra whitespace, and newlines before inference

---

🏗️ Architecture

User Browser
    │
    ▼
┌─────────────────────────────────────┐
│         FastAPI Application          │
│  ┌──────────────────────────────┐   │
│  │  POST /api/summarize         │   │
│  │  ├─ Input cleaning           │   │
│  │  ├─ Tokenization (seq ≤1024) │   │
│  │  └─ Beam Search (n=4)        │   │
│  └──────────────────────────────┘   │
│  ┌──────────────────────────────┐   │
│  │  GET /  → static/index.html  │   │
│  └──────────────────────────────┘   │
└──────────────┬──────────────────────┘
               │
               ▼
    ┌──────────────────────┐
    │  T5ForConditional    │
    │  Generation (CUDA)   │
    │  saved_summarizer_   │
    │  model/              │
    └──────────────────────┘

Model Specs:

Parameter	Value
Architecture	T5 Encoder-Decoder
Model Dimension (d_model)	768
Feed-Forward Dimension (d_ff)	3072
Attention Heads	12
Encoder Layers	12
Decoder Layers	12
Vocab Size	32,128
Input Max Length	1024 tokens
Output Length (default)	40–150 tokens
Decoding Strategy	Beam Search (4 beams, length_penalty=2.0)

---

📁 Project Structure

T5_Text_Summarizer/
│
├── main.py                        # FastAPI app — API routes & model loading
├── requirements.txt               # Python dependencies
├── .gitignore                     # Git ignore rules (excludes large model weights)
├── LICENSE                        # MIT License
├── README.md                      # Project documentation (this file)
│
├── static/                        # Frontend web app
│   ├── index.html                 # Main page — dual-panel UI (input / output)
│   ├── style.css                  # Dark glassmorphism CSS (Inter + JetBrains Mono)
│   └── script.js                  # API calls, slider logic, Three.js shader setup
│
└── saved_summarizer_model/        # Fine-tuned T5 model artifacts
    ├── config.json                # Model architecture config
    ├── generation_config.json     # Beam search & generation defaults
    ├── model.safetensors          # Model weights (~850 MB) — excluded from Git
    ├── tokenizer.json             # SentencePiece vocabulary & rules
    └── tokenizer_config.json      # Tokenizer settings & special tokens

Note: model.safetensors (~850 MB) is excluded from this repository via .gitignore. Download or train the model locally (see Training).

---

🚀 Quick Start

Prerequisites

• Python 3.9+
• NVIDIA GPU with CUDA drivers installed (required — CPU inference is disabled)
• Git

1. Clone the Repository

git clone https://github.com/nisargpatel1906/T5_Text_Summarizer.git
cd T5_Text_Summarizer

2. Create a Virtual Environment

python -m venv venv
# Windows
venv\Scripts\activate
# macOS/Linux
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

For GPU support, install the CUDA-compatible version of PyTorch:

pip install torch --index-url https://download.pytorch.org/whl/cu121

4. Place the Trained Model

Ensure your fine-tuned model files are available at:

saved_summarizer_model/
├── config.json
├── generation_config.json
├── model.safetensors      ← required (not in Git)
├── tokenizer.json
└── tokenizer_config.json

5. Run the Application

python main.py

The server will start at http://127.0.0.1:8000

Open the URL in your browser to access the web interface.

---

🛠️ API Reference

POST /api/summarize

Summarizes the provided input text using the fine-tuned T5 model.

Request Body:

{
  "text": "Your article or document text here...",
  "max_length": 150,
  "min_length": 40
}

Field	Type	Default	Description
text	string	—	The input text to summarize (required)
max_length	integer	150	Maximum token length of the generated summary
min_length	integer	40	Minimum token length of the generated summary

Success Response (200 OK):

{
  "summary": "A concise and coherent summary of the input text."
}

Error Responses:

Status	Reason
400	Input text is empty
500	Model not loaded or inference error

GET /

Serves the main index.html web interface.

Interactive Docs

FastAPI auto-generates interactive API documentation:

• Swagger UI: http://127.0.0.1:8000/docs
• ReDoc: http://127.0.0.1:8000/redoc

---

🧪 Training the Model

The model was fine-tuned using the Hugging Face Trainer API on a summarization dataset. The training notebook is included in the repository:

📓 text_summarizer.ipynb — Complete training pipeline including:

• Dataset loading and preprocessing
• T5 tokenization with summarize: prefix
• TrainingArguments configuration
• Evaluation using ROUGE metrics
• Model checkpointing and saving

Key Training Details:

• Base model: t5-base
• Task prefix: "summarize: "
• Input truncation: 1024 tokens
• Target truncation: 128 tokens
• Total epochs: 10
• Decoding: Beam search with num_beams=4, length_penalty=2.0, early_stopping=True

⚠️ Checkpoint Note: The model used in this application is taken from checkpoint-5500. Although the model was trained for all 10 epochs, the best validation performance and summary quality were achieved at step 5500 (mid-training). Beyond checkpoint 5500, while training continued through all 10 epochs, the model showed signs of overfitting with diminishing returns on validation metrics. Using checkpoint-5500 provides optimal generalization and the highest quality summaries on unseen text.

---

🖥️ Web Interface

The frontend is a single-page application with:

• Dual-panel layout — Source text input (left) and AI output (right)
• Three.js animated background — WebGL shader-based dark animated background
• Glassmorphism panels — Frosted-glass card design with subtle glow effects
• Word count indicator — Live word count as you type
• Loading skeleton — Shimmer animation while awaiting inference
• Copy to clipboard — One-click copy button on generated summary
• Fonts: Inter + JetBrains Mono from Google Fonts

---

⚙️ Dependencies

Package	Purpose
fastapi	Web framework for the REST API
uvicorn	ASGI server for running FastAPI
pydantic	Request/response data validation
transformers	T5 model & tokenizer (Hugging Face)
torch	PyTorch for GPU inference

---

🔭 Future Goals

The following improvements are planned for upcoming versions of this project:

#	Goal	Description
1	🧠 Upgrade to T5-Large / FLAN-T5	Replace t5-base with t5-large or Google's flan-t5-large for better factual accuracy and fluency in summaries
2	📂 Multi-File Batch Summarization	Allow users to upload multiple .txt or .pdf files at once and download a combined summary report
3	🌍 Multi-Language Support	Add support for summarizing text in multiple languages (Hindi, French, Spanish) using multilingual T5 (mT5)
4	📊 ROUGE Score Display	Show real-time ROUGE-1, ROUGE-2, and ROUGE-L scores alongside each generated summary so users can gauge quality
5	🗂️ Summary History & Export	Save past summaries in a local session log with the ability to export them as .txt or .pdf files

---

🤝 Contributing

Contributions, issues, and feature requests are welcome!

1. Fork this repository
2. Create your feature branch: git checkout -b feature/amazing-feature
3. Commit your changes: git commit -m 'feat: add amazing feature'
4. Push to the branch: git push origin feature/amazing-feature
5. Open a Pull Request

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for full details.

---

👤 Author

Nisarg Patel

• GitHub: @nisargpatel1906

---

_{Built with ❤️ using T5, FastAPI & PyTorch · Fine-tuned for abstractive text summarization}