mmBERT: ModernBERT goes Multilingual

mmBERT is a massively multilingual encoder trained on over 3T tokens across 1,800+ languages, achieving state-of-the-art results with faster inference than prior models. It extends ModernBERT with novel components and a three-phase training schedule to improve learning for low-resource languages, using progressively sampled, broader data.

Source

HuggingFace Blog

Ingéré

09 SEPT. 2025 · 19:10

Score édito

4.0 / 5