Borealis 1B Instruct (Preview) Overview

NbAiLab/borealis-1b-instruct-preview is an experimental 1-billion parameter instruction-tuned model from NbAiLab, released for early testing and feedback. Based on the google/gemma-3-1b-it architecture, this model has been fine-tuned solely on textual instructions using the NbAiLab/aurora-sft-2512 dataset.

Key Characteristics

• Experimental Status: This is a preview model, meaning it is pre-release quality and intended for early evaluation. Outputs may be unstable and prone to hallucination.
• Norwegian Focus: Specifically designed for Norwegian-centric assistant tasks, including drafting, summarization, Q&A, and light reasoning.
• Safety Disclaimer: Crucially, this model has not been safety-aligned. It may produce harmful, biased, or inappropriate content and should not be used in safety-critical applications without additional mitigations.
• Format Availability: Provided in Transformers (safetensors), GGUF (q8_0, f16, bf16), and MLX (32-bit and 8-bit quantized) formats, supporting various deployment environments including Apple Silicon.

Good for

• Early assessment of model behavior and quality with Norwegian language (Bokmål and Nynorsk).
• Evaluating Norwegian writing style and content generation.
• Testing assistant-style tasks in a Norwegian context.
• Providing feedback on an experimental, unaligned model for future development.