beto

Why we included this project

Teams building Spanish-language NLP pipelines have long needed a BERT checkpoint trained on Spanish itself, not a multilingual model spread thin across many other languages. BETO is that: a BERT-Base sized model trained with Whole Word Masking on a large Spanish corpus. The README's benchmark tables on POS tagging, named entity recognition, and other Spanish tasks show it typically matching or beating the best multilingual BERT results, which is useful evidence before you adopt it. Both uncased and cased checkpoints are on Hugging Face, so loading either takes a couple of lines with the Transformers library rather than manual conversion. The one thing to verify before production is licensing: the project is released under CC BY 4.0, and the authors flag that the training data licenses may not permit commercial use.