hmnshudhmn24/emotion-rewrite-transformer

🎭 Emotional Tone Rewriter

A Transformer-based NLP model that rewrites sentences into different emotional tones such as:

• friendly
• angry
• sad
• corporate
• romantic
• sarcastic
• motivational
• formal
• cheerful

Built using T5-small, this project includes a complete pipeline: preprocessing, training, evaluation, inference, FastAPI server, and Gradio UI.

🚀 Features

✓ Rewrite any sentence into a selected emotional tone
✓ Fully reproducible training pipeline (PyTorch + HuggingFace)
✓ Clean dataset preprocessing (CSV → JSONL)
✓ Gradio UI for interactive demos
✓ FastAPI server for deployment
✓ Evaluation notebook with ROUGE metrics
✓ Apache 2.0 licensed (commercial-friendly)
✓ Production-grade project structure

📁 Project Structure

emotional-tone-rewriter/
├── data/
│   ├── raw/
│   ├── processed/
│   └── README.md
├── src/
│   ├── config.py
│   ├── dataset_preprocessing.py
│   ├── train.py
│   ├── evaluate.py
│   ├── inference.py
│   └── model_utils.py
├── app/
│   ├── api.py
│   └── ui.py
├── model/
├── notebooks/
├── huggingface/
├── requirements.txt
├── LICENSE
└── README.md

📦 Installation

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

🔄 Dataset Preprocessing

python -m src.dataset_preprocessing --input data/raw/dataset_raw.csv --output data/processed/dataset_clean.jsonl

🏋️ Train Model

python -m src.train

🧪 Evaluate

python -m src.evaluate

🤖 Inference Example

from src.inference import rewrite_tone
print(rewrite_tone("I can't come today.", "sarcastic"))

🌐 FastAPI Server

uvicorn app.api:app --reload --port 7860

🎨 Gradio UI

python app/ui.py

📄 License

Apache License 2.0