Skip to content
TopicTracker
From HackerNewsView original
TranslationTranslation

Amalia – an open-source language model targeting European Portuguese

Amalia is an open-source language model specifically developed for European Portuguese, aiming to improve natural language processing capabilities for this language variant.

Background

- Amalia is an open-source large language model (LLM) built specifically for European Portuguese (the variant spoken in Portugal), not Brazilian Portuguese. Most major LLMs are trained mainly on English and treat Portuguese as an afterthought, usually defaulting to Brazilian usage, leaving European Portuguese underserved in grammar, vocabulary, and cultural context. - The model is hosted on Hugging Face, the main platform for sharing open-source AI models (like GitHub for machine learning), where anyone can download, use, or fine-tune them. - This matters because it reduces dependence on US tech giants (OpenAI, Google, Meta) and gives Portugal its own AI tool that can handle local needs — tax forms, legal language, news — properly. - Amalia is part of a growing wave of region-specific open-source LLMs, similar to Sabia (Brazilian Portuguese), as well as models for Catalan, Basque, and Japanese.