sdobson/tinystories-llama-15m

TinyStories Story Generator

An interactive Gradio Space for generating simple children's stories using a small Llama-architecture model trained on the TinyStories dataset.

About

This Space provides a chat-style interface to interact with a ~15M parameter language model that generates simple, coherent children's stories. The model uses vocabulary and concepts that a typical 3-4 year old would understand.

Features

• —Interactive story generation
• —Adjustable generation parameters (temperature, top-k, top-p, max length)
• —Example prompts to get started
• —Real-time generation
• —Clean, user-friendly interface

Model Details

• —Architecture: Llama 2
• —Parameters: ~15M
• —Layers: 6
• —Attention Heads: 6
• —Max Context Length: 256 tokens
• —Training Dataset: TinyStories

Usage

1. 1.Enter a story prompt (e.g., "Once upon a time, there was a...")
2. 2.Optionally adjust generation settings
3. 3.Click "Generate Story"
4. 4.Enjoy your AI-generated children's story!

Setup Instructions

To run this Space:

1. 1.Upload your trained model to HuggingFace Hub
2. 2.Update the MODEL_REPO variable in app.py with your model repository (format: username/model-name)
3. 3.Or set the MODEL_REPO environment variable in the Space settings

Local Development

pip install -r requirements.txt
python app.py

License

MIT License

Acknowledgments

• —Model architecture and training code adapted from llama2.c by Andrej Karpathy
• —Trained on the TinyStories dataset by Ronen Eldan and Yuanzhi Li
• —Based on the Llama 2 architecture by Meta AI