Summary 🔥
A micro-LLM (~12.1M parameters) trained on Dante Alighieris works and medieval Italian literature that runs entirely in the browser to generate infinite Dante-style text.
Architecture 🏗️
- Built on nanoGPT architecture: 15 transformer layers, 256 embedding dimension, 8 attention heads
- Two-phase training pipeline: pre-training on general Italian medieval literature, fine-tuning on Dantes Divine Comedy and related works
- BPE tokenization with 512-token vocabulary trained on the combined corpus
- Exported to ONNX for browser inference via ONNX Runtime Web
- Interactive web UI with minimalist medieval design (Cormorant Garamond font)
Why it matters 💡
- Demonstrates that sophisticated LLMs can run entirely client-side with no backend dependencies
- Combines modern deep learning techniques with classical literature in an innovative way
- Educational project showing the full ML pipeline: data preparation, training, optimization, and deployment