Show HN: Tinytowns, a small model to generate realistic sounding town names

Category: ai-ml

Tags: transformer, text-generation, educational, tiny-ml, names

Score: 5.5/10 (Innovation: 5, Technical: 5, Documentation: 8, Utility: 4)

Tinytowns is a tiny transformer model trained on 142k place names to generate realistic-sounding town names. It's notable for its extreme simplicity and transparency—the smallest model has only 4,572 weights—making it an educational tool for understanding transformer internals. The project includes a JavaScript library for easy integration and detailed documentation of the training process and results.

Target audience: machine learning learners, game developers, writers, and hobbyists

Repository: https://quirkyagents.com/wizard/projects/tinytowns/ · Python · MIT

View on Hacker News