Text Generator

A web application that generates human-like text using OpenAI's GPT-2 large language model. This project implements a Flask-based web interface where users can input prompts and receive AI-generated text completions.

Features

• Text generation using GPT-2 large model from HuggingFace
• Web interface with Bootstrap styling
• Interactive user experience with form submission
• Text generation with configurable parameters
• Responsive design

Technologies Used

• Python: Core programming language
• Flask: Web framework for serving the application
• HuggingFace Transformers: For accessing pre-trained GPT-2 model
• PyTorch: Deep learning framework that powers the model
• Bootstrap: Front-end styling and responsive design
• HTML/CSS: Web interface design

Installation

1. Clone the repository:

   git clone https://github.com/Shams261/Text-Generator.git
   cd Text-Generator
   

2. Install the required packages:

   pip install flask torch transformers
   

Usage

1. Run the Flask application:

   python app.py
   

2. Open your web browser and navigate to:

   http://127.0.0.1:5000/
   

3. Enter a prompt in the input field and click "Generate Text" to see the AI-generated completion.

Project Structure

Text-Generator/
├── app.py                   # Flask application
├── TextGenerationGpt2Huggingface.ipynb  # Jupyter notebook for model exploration
├── models/                  # Saved model files
│   └── tokenizer.pickle     # Saved tokenizer
├── templates/               # HTML templates
│   └── index.html           # Main web interface
└── README.md                # Project documentation

How It Works

The application uses a pre-trained GPT-2 large model from HuggingFace's Transformers library. When a user submits a prompt, the application:

1. Tokenizes the input text
2. Passes it through the GPT-2 model
3. Generates text completion with beam search and n-gram repetition prevention
4. Returns the generated text to the user interface

Model Parameters

The text generation uses the following parameters:

• max_length: 300 tokens
• num_beams: 5 (beam search for better quality)
• no_repeat_ngram_size: 2 (prevents repetition of n-grams)
• early_stopping: True (stops generation when all beams reach EOS)

Future Improvements

• Add temperature control for text generation
• Implement different model size options
• Add save/export functionality for generated text
• Incorporate user feedback mechanism
• Add more advanced text formatting options

License

This project is open source and available under the MIT License.

Acknowledgements

• OpenAI for creating the GPT-2 model
• HuggingFace for providing access to pre-trained models
• Flask for the web framework

---

Created by Shams Tabrez