Gujarati AI Poetry Generator

🌟 Introduction

This project explores the intersection of Artificial Intelligence and Satsang literature by generating Gujarati poetry inspired by the works of great poets like Nishkulanand Swami. Using AI models, this tool transforms English prompts into meaningful, poetic Gujarati verses.

✨ Features

Gujarati BPE Tokenizer: Custom-built tokenizer for efficient text generation.
Fine-tuned GPT Model: Leveraging GPT-J/GPT-2 for Gujarati poetry generation.
Dataset: Based on structured English-Gujarati poetry pairs.
Use Cases in Satsang: AI-driven tools for kirtan composition, scripture analysis, and poetic expressions.

🛠️ Installation

# Clone the repository
git clone https://github.com/yourusername/gujarati-ai-poetry.git
cd gujarati-ai-poetry

# Install dependencies
pip install -r requirements.txt

📜 Dataset Structure

The training dataset follows this format:

{
    "gujarati": [
        {
            "title_gu": "પ્રકરણ ૧: મંગળાચરણ",
            "title_trans": "Prakaran 1: Mangaḷācharaṇ",
            "target": "મંગલમૂર્તિ મહાપ્રભુ, શ્રીસહજાનંદ સુખરૂપ...",
            "source": "The divine Lord is the embodiment of joy and beauty..."
        }
    ]
}

🚀 Training & Fine-Tuning

To train the model:

python train.py --model gpt-j --dataset cleaned_final_data.json

To fine-tune GPT-2 from scratch:

python train_gpt2.py --epochs 5 --batch_size 8

🔥 Inference Example

from model import generate_poetry
prompt = "The divine Lord is the embodiment of joy and beauty."
response = generate_poetry(prompt)
print(response)

🎯 Future Enhancements

Improving rhyme and meter matching.
Expanding dataset with more scriptures & kirtans.
Deploying a web-based interface for real-time poetry generation.

🌟 Contribute

Feel free to open issues and pull requests! Your contributions can help refine AI-driven Gujarati poetry generation.

📜 License

This project is licensed under the MIT License.

🔗 For more AI & Satsang insights, connect with me on LinkedIn!

shivampadmani
/

nishkulanandAI