AMÁLIA and the future of European Portuguese LLMs | Hacker News

Hacker News debated how to train LLMs that truly reflect European Portuguese.

One side argued for collecting and documenting local datasets and cultural artifacts like AMÁLIA and Miranda de Douro materials.

Others said broader multilingual or Brazilian Portuguese-heavy base training, followed by European Portuguese fine-tuning, may be more practical.

The discussion underscored the difficulty of building high-quality models for a small language variant with limited text data.