Norwegian NER (nb-bert-base fine-tuned)
This is a solid, practical Named Entity Recognition (NER) model for Norwegian Bokmål and Nynorsk, built by fine-tuning NbAiLab/nb-bert-base on a combined dataset (NorNE + WikiANN). It’s a good fit for “classic NLP plumbing” tasks—pulling people, organizations, and place names out of Norwegian text—where you want something lightweight and predictable rather than a general-purpose chat model.
The model card includes clear evaluation numbers (F1 ≈ 0.93, with precision/recall around 0.93 as well) and documents the supported entity types: PER, ORG, LOC, and MISC. To try it quickly, you can run it with the transformers pipeline("ner", ...) API and an aggregation strategy like simple to merge subword tokens. If you’re building search, analytics, redaction, or enrichment for Norwegian content, this is the kind of model that can slot in cleanly.
The training data is a fairly substantial combined set (tens of thousands of samples) and the card is refreshingly specific about data sources and training choices. It’s also published under a permissive CC BY 4.0 license, which makes it easier to adopt in downstream projects.
Quick stats from the listing feed: pipeline: token-classification · 148 downloads.
Source listing: https://huggingface.co/models?sort=modified