Bible LLM

A lightweight, PyTorch-based character-level Transformer Language Model designed for Biblical-style text generation. Optimized for CPU training and featuring a hybrid search capability for exact verse retrieval.

Features

• Hybrid Search: Automatically detects Bible references (e.g., "Genesis 1:1") and returns exact text from the dataset before falling back to AI generation.
• Multi-Version Support: Specify different Bible versions (KJV, WEB, etc.) for training and generation.
• Resumable Training: Smart checkpoints allow you to pause and resume training sessions without losing progress.
• FastAPI Integration: A fully functional REST API for remote training and text generation.

Setup

1. Create a virtual environment (optional but recommended):

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

2. Install dependencies:

   pip install -r requirements.txt

Datasets

Place your Bible text files in the datasets/ directory. Files must be named {version}.txt (e.g., kjv.txt, web.txt).

Usage

1. Training (via CLI)

To train the model on a specific version:

python src/train.py --version kjv --max_iters 5000

• Use --no-resume to start training from scratch instead of continuing from a previous session.
• Weights are saved to model_{version}.pt.

2. Generation (via CLI)

Generate text or look up a verse:

python src/generate.py --version kjv --prompt "Genesis 1:1"

• If the prompt is a citation, it returns the exact verse.
• Otherwise, it uses the AI model to generate text.

3. API (FastAPI)

Run the server:

python main.py

Generate Text:

curl -X POST "http://127.0.0.1:8000/generate" \
     -H "Content-Type: application/json" \
     -d '{"prompt": "In the beginning", "version": "kjv", "max_tokens": 100}'

Train Model:

curl -X POST "http://127.0.0.1:8000/train" \
     -H "Content-Type: application/json" \
     -d '{"version": "kjv", "max_iters": 500, "resume": true}'

Utility Endpoints:

• GET /versions: List available datasets.
• GET /status: Check current training progress.
• http://127.0.0.1:8000/docs: Interactive Swagger UI.

4. Contribution

To contribute, fork this repo, make your changes, and submit a pull request.

Thank you.