An LLM built from scratch. Pure NumPy. No frameworks, just math. A real character-level transformer you can train and run.
Real trained transformer running in your browser. Type a prompt:
Text → tokens → embeddings → 2 transformer layers (attention + MLP) → next-character probabilities. Full code in the GitHub repo.