NbAiLab/borealis-27b

Borealis 27B: Norwegian-Centric Instruction-Tuned Model

NbAiLab/borealis-27b is a 27-billion parameter instruction-tuned model developed by the National Library of Norway (NbAiLab). It is built upon the Gemma 3 family, specifically fine-tuned from google/gemma-3-27b-it, with a focus on Norwegian language tasks.

Key Capabilities

• Norwegian Language Specialization: Optimized for instruction following in Norwegian, including Bokmål and Nynorsk.
• Unique Training Data: Incorporates a limited subset of copyright-protected Norwegian press material, enabling the model to generate news article titles and ingress texts.
• Assistant-Style Tasks: Proficient in drafting, summarization, and question answering for Norwegian content.
• Safety Alignment: Utilizes "prompt baking" and weighted merging of SFT and aligned models to balance quality, usefulness, and safer behavior.

Good For

• Norwegian Language Applications: Ideal for developers and researchers working on AI solutions requiring deep understanding and generation of Norwegian text.
• Content Generation: Useful for creating Norwegian news article titles, ingress texts, and other structured content.
• Assistant Development: Suitable for building Norwegian-centric virtual assistants, chatbots, and summarization tools.
• Writing Assessment: Can be used for assessing and improving Norwegian writing style and quality.