Assignment 4: Sentence Style Transfer (Formal ↔ Informal)

Objective

Build a model/pipeline to convert an informal sentence into a formal one and vice versa.

Models Used

• T5 Fine-tuned Model: prithivida/informal_to_formal_styletransfer from Hugging Face Transformers for informal-to-formal conversion.
• gpt-4o-mini: Use prompting with gpt-4o-mini for sentence style transfer.

Project Setup & Usage

1. Clone the Repository

Clone or download this repository to your local machine.

2. Create and Activate a Virtual Environment (Recommended)

Navigate to the project directory and run:

python3 -m venv venv
source venv/bin/activate

3. Install Dependencies

With the virtual environment activated, install all required packages:

pip install --upgrade pip
pip install torch transformers openai python-dotenv nltk streamlit

4. Set Up OpenAI API Key

Create a .env file in the project root with the following content:

OPENAI_API_KEY=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Replace xxxxxxxx... with your actual OpenAI API key.

5. Run the Notebook

Open sentence_style_transfer.ipynb in Jupyter or VS Code. Make sure the kernel/interpreter is set to the Python from your venv (see VS Code's top-right kernel picker).

Run all cells sequentially. The notebook will:

• Import libraries
• Load the T5 model
• Provide an inference pipeline
• Evaluate BLEU score
• Allow you to test custom sentences

6. Run the Streamlit Web App

To launch the web interface:

streamlit run style_transfer_streamlit.py

The app will open in your browser at http://localhost:8501

---

Requirements

• Python 3.8+
• transformers, datasets, torch, evaluate, pandas, numpy

UI Demo

Run the UI script for a simple web interface:

python style_transfer_ui.py

---