NbAiLab/borealis-270m

Borealis 270M Overview

NbAiLab/borealis-270m is a 270-million parameter instruction-tuned model developed by the National Library of Norway (NbAiLab). It is part of the Borealis family, built upon the Gemma 3 architecture, specifically fine-tuned from google/gemma-3-270m-it.

Key Capabilities

• Norwegian Language Focus: Optimized for Norwegian-centric instruction following, including Bokmål and Nynorsk.
• Specialized Data Integration: This is the first Borealis release to incorporate data from an agreement with Norwegian rights-holder organizations, specifically used to teach the model to generate news article titles and ingress texts.
• Instruction Following: Trained on the NbAiLab/aurora-sft dataset for assistant-style behavior, writing, summarization, and question answering in Norwegian.
• Safety Alignment: Utilizes a technique called prompt baking to distill system prompt behavior directly into model weights, balancing quality, usefulness, and safer outputs.

Good For

• Norwegian Assistant Tasks: Ideal for drafting, summarization, question answering, and light reasoning in Norwegian contexts.
• Norwegian Writing Improvement: Useful for assessing and enhancing Norwegian writing style and quality.
• Language Coverage Evaluation: Suitable for evaluating model behavior and language coverage across Norwegian, Bokmål, and Nynorsk.