File size: 5,031 Bytes
b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 7843c1b b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 b59c3e9 1b49816 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 |
---
license: gemma
datasets:
- NbAiLab/aurora-sft-2512-filtered
language:
- 'no'
- nb
- nn
base_model:
- google/gemma-3-27b-it
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- conversational
- instruct
- experimental
---
# Borealis 27B Instruct (Preview)
Release: Jan 21st, 2026.
## Model summary
**NbAiLab/borealis-27b-instruct-preview** is a **27B-parameter** instruction-tuned **preview** model intended for early testing and feedback. It is an **experiment** and should be treated as pre-release quality.
This model is based on [**google/gemma-3-27b-it**](https://huggingface.co/google/gemma-3-27b-it), and fine-tuned on textual instructions only.
| Size | Transformers | GGUF (quants: `q8_0`, `f16`, `bf16`) | MLX |
|---:|---|---|---|
| 270M | [NbAiLab/borealis-270m-instruct-preview](https://huggingface.co/NbAiLab/borealis-270m-instruct-preview) | [NbAiLab/borealis-270m-instruct-preview-gguf](https://huggingface.co/NbAiLab/borealis-270m-instruct-preview-gguf) | [NbAiLab/borealis-270m-instruct-preview-mlx](https://huggingface.co/NbAiLab/borealis-270m-instruct-preview-mlx) (32-bit)<br>[NbAiLab/borealis-270m-instruct-preview-mlx-8bits](https://huggingface.co/NbAiLab/borealis-270m-instruct-preview-mlx-8bits) (8-bit) |
| 1B | [NbAiLab/borealis-1b-instruct-preview](https://huggingface.co/NbAiLab/borealis-1b-instruct-preview) | [NbAiLab/borealis-1b-instruct-preview-gguf](https://huggingface.co/NbAiLab/borealis-1b-instruct-preview-gguf) | [NbAiLab/borealis-1b-instruct-preview-mlx](https://huggingface.co/NbAiLab/borealis-1b-instruct-preview-mlx) (32-bit)<br>[NbAiLab/borealis-1b-instruct-preview-mlx-8bits](https://huggingface.co/NbAiLab/borealis-1b-instruct-preview-mlx-8bits) (8-bit) |
| 4B | [NbAiLab/borealis-4b-instruct-preview](https://huggingface.co/NbAiLab/borealis-4b-instruct-preview) | [NbAiLab/borealis-4b-instruct-preview-gguf](https://huggingface.co/NbAiLab/borealis-4b-instruct-preview-gguf) | [NbAiLab/borealis-4b-instruct-preview-mlx](https://huggingface.co/NbAiLab/borealis-4b-instruct-preview-mlx) (32-bit)<br>[NbAiLab/borealis-4b-instruct-preview-mlx-8bits](https://huggingface.co/NbAiLab/borealis-4b-instruct-preview-mlx-8bits) (8-bit) |
| 12B | [NbAiLab/borealis-12b-instruct-preview](https://huggingface.co/NbAiLab/borealis-12b-instruct-preview) | [NbAiLab/borealis-12b-instruct-preview-gguf](https://huggingface.co/NbAiLab/borealis-12b-instruct-preview-gguf) | [NbAiLab/borealis-12b-instruct-preview-mlx](https://huggingface.co/NbAiLab/borealis-12b-instruct-preview-mlx) (32-bit)<br>[NbAiLab/borealis-12b-instruct-preview-mlx-8bits](https://huggingface.co/NbAiLab/borealis-12b-instruct-preview-mlx-8bits) (8-bit) |
| 27B | [NbAiLab/borealis-27b-instruct-preview](https://huggingface.co/NbAiLab/borealis-27b-instruct-preview) | [NbAiLab/borealis-27b-instruct-preview-gguf](https://huggingface.co/NbAiLab/borealis-27b-instruct-preview-gguf) | [NbAiLab/borealis-27b-instruct-preview-mlx](https://huggingface.co/NbAiLab/borealis-27b-instruct-preview-mlx) (32-bit)<br>[NbAiLab/borealis-27b-instruct-preview-mlx-8bits](https://huggingface.co/NbAiLab/borealis-27b-instruct-preview-mlx-8bits) (8-bit) |
## Training data
Supervised fine-tuning (SFT) uses **NbAiLab/aurora-sft-2512** (not released yet).
## ⚠️ Safety / alignment disclaimer (important)
This is a **preview experiment** and **has not been safety-aligned yet**. The model may produce **harmful, biased, or insensitive** outputs (including content that is offensive, unsafe, or inappropriate). Do not use it for safety-critical or high-stakes applications, and add your own safety mitigations if deploying.
## Intended use
- Norwegian-centric assistant-style tasks (e.g., drafting, summarization, Q&A, light reasoning).
- Assesstment of Norwegian writing style and quality.
- Early evaluation of behavior, language coverage (Norwegian / Bokmål / Nynorsk), and quality.
## Limitations
- Preview quality; outputs may be unstable and may hallucinate.
- Not aligned for safety; may follow harmful instructions or generate problematic content (see disclaimer above).
## Weights & formats
### Transformers (original)
- **NbAiLab/borealis-27b-instruct-preview** (safetensors).
### GGUF quantizations
Available in [**NbAiLab/borealis-27b-instruct-preview-gguf**](https://huggingface.co/NbAiLab/borealis-27b-instruct-preview-gguf):
- `model-q8_0.gguf`
- `model-f16.gguf`
- `model-bf16.gguf`
Use:
```bash
ollama run hf.co/NbAiLab/borealis-27b-instruct-preview-gguf:BF16
```
### MLX (Apple Silicon)
Available in [**NbAiLab/borealis-27b-instruct-preview-mlx**](https://huggingface.co/NbAiLab/borealis-27b-instruct-preview-mlx) and quantized to [8 bits](https://huggingface.co/NbAiLab/borealis-27b-instruct-preview-mlx-8bits).
Use:
```bash
# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "NbAiLab/borealis-27b-instruct-preview-mlx"
```
## Acknowledgements
Thanks to the **Gemma** team at Google for releasing Gemma 3 and to everyone contributing feedback on this preview. |