NbAiLab/borealis-27b

Borealis 27B: Norwegian-Centric Instruction Model

NbAiLab/borealis-27b is a 27 billion parameter instruction-tuned model developed by the National Library of Norway (Nasjonalbiblioteket, NB). It is built upon the Gemma 3 architecture, specifically fine-tuned from google/gemma-3-27b-it.

Key Differentiators & Capabilities

• Norwegian Language Focus: This model is specifically instruction-tuned for Norwegian-centric tasks, including assistant behavior, writing, summarization, and question answering.
• Unique Training Data: It incorporates a limited subset of data from an agreement with Norwegian rights-holder organizations, including copyrighted press material, to enhance its understanding and generation of Norwegian news article titles and ingress texts. The SFT dataset, NbAiLab/aurora-sft, includes 10k tasks derived from this material.
• Safety Alignment: Utilizes "prompt baking" and weighted merging of SFT and aligned models to balance quality, usefulness, and safer behavior.

Intended Use Cases

• Norwegian-centric assistant-style tasks, such as drafting, summarization, and Q&A.
• Assessment and improvement of Norwegian writing style and quality.
• Evaluation of behavior and language coverage for Norwegian, Bokmål, and Nynorsk.

Limitations

• May produce incorrect or biased outputs.
• Not recommended for safety-critical applications without additional safeguards.
• Performance outside Norwegian and English use cases is not fully characterized.