Instructions to use NbAiLab/borealis2-26b-a4b-preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use NbAiLab/borealis2-26b-a4b-preview with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="NbAiLab/borealis2-26b-a4b-preview")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("NbAiLab/borealis2-26b-a4b-preview")
model = AutoModelForMultimodalLM.from_pretrained("NbAiLab/borealis2-26b-a4b-preview", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use NbAiLab/borealis2-26b-a4b-preview with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "NbAiLab/borealis2-26b-a4b-preview"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NbAiLab/borealis2-26b-a4b-preview",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/NbAiLab/borealis2-26b-a4b-preview

SGLang

How to use NbAiLab/borealis2-26b-a4b-preview with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "NbAiLab/borealis2-26b-a4b-preview" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NbAiLab/borealis2-26b-a4b-preview",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "NbAiLab/borealis2-26b-a4b-preview" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NbAiLab/borealis2-26b-a4b-preview",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use NbAiLab/borealis2-26b-a4b-preview with Docker Model Runner:
```
docker model run hf.co/NbAiLab/borealis2-26b-a4b-preview
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Borealis 2 26B-A4B (Preview)

Release: August 21st, 2026.

Model summary

NbAiLab/borealis2-26b-a4b-preview is a 26B-parameter (4B active) instruction-tuned preview model intended for early testing and feedback. It is an experiment and should be treated as pre-release quality.

The model starts from google/gemma-4-26B-A4B-it, continued pre-trained on Aurora 2604 data, and then supervised fine-tuned for two epochs on textual instructions from NbAiLab/aurora-sft-2606. SFT loss is calculated on assistant responses only.

Training data

Continuous pre-training uses aurora, a mix of sources both internal and external, including license materials under agreements. Supervised fine-tuning (SFT) uses NbAiLab/aurora-sft-2606. Examples use the Gemma 4 chat template; user and system tokens are masked from the training loss.

The aurora dataset finished building in April 2026 and contained a total of 47B whitespace-separated words with a knowledge cutoff of January 1st 2025.

Sources

Word counts

Source\Language	Total	nob	nno	dan	swe	eng	fao	isl	sme	non	fin	smn	sma	smj	sms	fkv	sju	other
Total	47,588,939,435	24,310,675,244	11,121,117,598	4,635,647,954	2,466,930,277	2,418,581,785	1,534,499,832	618,403,302	136,625,254	130,821,960	115,529,546	26,099,190	25,054,850	19,392,470	11,725,220	10,308,610	42,900	7,483,443
hplt-3.0	15,020,676,845	7,387,980,881	1,650,457,860	3,406,653,847	1,945,721,769	241	386,912,020	161,832,408	509,882	39,453,060	37,167,967	633,100	46,480	152,950	15,170	3,139,210	-	-
fineweb-2	14,616,279,115	9,586,371,459	3,739,975,210	-	-	91	814,239,570	273,740,521	52,936,570	47,996,140	62,367,474	11,890,230	9,333,600	5,615,090	6,946,090	4,826,690	40,380	-
nb-newspapers-mbl	6,403,459,654	3,222,919,898	3,179,351,370	36,981	12,848	205,425	-	539	76,101	1,090	102	-	551,030	27,710	-	276,560	-	-
finepdfs	5,885,083,486	3,031,393,119	1,161,632,350	757,427,249	416,681,332	3,914	280,421,310	181,277,900	16,222,925	13,795,380	15,306,017	2,480,840	1,423,800	4,614,610	2,057,710	345,030	-	-
maalfrid-crawl	994,747,286	492,399,964	482,117,030	565,919	39,357	9,450,493	-	-	5,264,213	188,860	-	-	2,787,440	1,902,390	-	31,620	-	-
swallow-code-v2	848,043,875	42,982	12,429	16,500,328	429,047	827,993,226	2,432	10,507	424	-	76,313	-	-	-	-	-	-	2,976,187
nb-books-ocr	817,147,445	220,740,790	459,992,080	96,760,244	448,854	1,267,865	601,140	-	7,256,662	29,052,580	-	-	714,070	313,160	-	-	-	-
megamath	795,092,965	3,201	6,419	858,060	201,115	789,418,493	-	-	12,292	-	86,129	-	-	-	-	-	-	4,507,256
fineweb-2-hq	439,744,102	66	-	344,871,624	94,872,412	-	-	-	-	-	-	-	-	-	-	-	-	-
finewiki	425,260,298	83,273,602	210,355,610	11,313,266	8,520,965	97,669,581	8,341,750	1,526,141	206,149	45,590	525,544	3,482,100	-	-	-	-	-	-
maalfrid-government	382,884,652	175,367,608	204,521,790	158,061	1,319	602,180	-	1,486	1,204,748	-	-	-	453,730	236,870	-	336,860	-	-
fineweb-edu	358,894,808	-	-	-	-	358,807,998	-	-	-	86,810	-	-	-	-	-	-	-	-
fineweb-hq	290,431,291	1,036	-	-	228	290,244,437	-	-	-	185,590	-	-	-	-	-	-	-	-
giellalt	112,384,962	3,616,964	2,479,390	20,829	1,031	-	43,981,610	13,800	34,305,758	16,860	-	7,612,920	9,744,700	6,529,690	2,706,250	1,352,640	2,520	-
parlamint	69,692,697	69,256,627	436,070	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
kudos	59,909,727	31,938,952	27,698,540	-	-	191,476	-	-	80,759	-	-	-	-	-	-	-	-	-
project-gutenberg	42,718,920	-	-	-	-	42,718,920	-	-	-	-	-	-	-	-	-	-	-	-
saami-web	18,548,893	-	-	-	-	122	-	-	18,548,771	-	-	-	-	-	-	-	-	-
lovdata	5,748,677	3,702,797	2,045,880	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
patents	1,132,943	1,132,671	-	272	-	-	-	-	-	-	-	-	-	-	-	-	-	-
nb-digimanus	1,056,794	532,627	35,570	481,274	-	7,323	-	-	-	-	-	-	-	-	-	-	-	-

Before upsampling

Source\Language	Total	nob	dan	swe	eng	nno	isl	fao	fin	sme	non	smn	sma	smj	sms	fkv	sju	other
Total	35,903,335,353	24,310,675,244	4,635,647,954	2,466,930,277	2,418,581,785	1,112,128,723	618,403,302	153,452,172	115,529,546	42,158,387	13,082,196	2,609,919	2,505,485	1,939,247	1,172,522	1,030,861	4,290	7,483,443
hplt-3.0	13,147,596,474	7,387,980,881	3,406,653,847	1,945,721,769	241	165,045,786	161,832,408	38,691,202	37,167,967	158,376	3,945,306	63,310	4,648	15,295	1,517	313,921	-	-
fineweb-2	10,402,853,593	9,586,371,459	-	-	91	373,997,521	273,740,521	81,423,957	62,367,474	16,287,748	4,799,614	1,189,023	933,360	561,509	694,609	482,669	4,038	-
finepdfs	4,553,779,407	3,031,393,119	757,427,249	416,681,332	3,914	116,163,235	181,277,900	28,042,131	15,306,017	5,012,773	1,379,538	248,084	142,380	461,461	205,771	34,503	-	-
nb-newspapers-mbl	3,541,220,733	3,222,919,898	36,981	12,848	205,425	317,935,137	539	-	102	24,164	109	-	55,103	2,771	-	27,656	-	-
swallow-code-v2	848,043,875	42,982	16,500,328	429,047	827,993,226	12,429	10,507	2,432	76,313	424	-	-	-	-	-	-	-	2,976,187
megamath	795,092,965	3,201	858,060	201,115	789,418,493	6,419	-	-	86,129	12,292	-	-	-	-	-	-	-	4,507,256
maalfrid-crawl	552,781,457	492,399,964	565,919	39,357	9,450,493	48,211,703	-	-	-	1,622,990	18,886	-	278,744	190,239	-	3,162	-	-
fineweb-2-hq	439,744,102	66	344,871,624	94,872,412	-	-	-	-	-	-	-	-	-	-	-	-	-	-
nb-books-ocr	370,533,410	220,740,790	96,760,244	448,854	1,267,865	45,999,208	-	60,114	-	2,248,354	2,905,258	-	71,407	31,316	-	-	-	-
fineweb-edu	358,816,679	-	-	-	358,807,998	-	-	-	-	-	8,681	-	-	-	-	-	-	-
fineweb-hq	290,264,260	1,036	-	228	290,244,437	-	-	-	-	-	18,559	-	-	-	-	-	-	-
finewiki	225,114,542	83,273,602	11,313,266	8,520,965	97,669,581	21,035,561	1,526,141	834,175	525,544	62,938	4,559	348,210	-	-	-	-	-	-
maalfrid-government	197,068,531	175,367,608	158,061	1,319	602,180	20,452,179	1,486	-	-	382,952	-	-	45,373	23,687	-	33,686	-	-
parlamint	69,300,234	69,256,627	-	-	-	43,607	-	-	-	-	-	-	-	-	-	-	-	-
project-gutenberg	42,718,920	-	-	-	42,718,920	-	-	-	-	-	-	-	-	-	-	-	-	-
kudos	34,922,645	31,938,952	-	-	191,476	2,769,854	-	-	-	22,363	-	-	-	-	-	-	-	-
giellalt	21,681,541	3,616,964	20,829	1,031	-	247,939	13,800	4,398,161	-	10,586,259	1,686	761,292	974,470	652,969	270,625	135,264	252	-
saami-web	5,736,876	-	-	-	122	-	-	-	-	5,736,754	-	-	-	-	-	-	-	-
lovdata	3,907,385	3,702,797	-	-	-	204,588	-	-	-	-	-	-	-	-	-	-	-	-
patents	1,132,943	1,132,671	272	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
nb-digimanus	1,024,781	532,627	481,274	-	7,323	3,557	-	-	-	-	-	-	-	-	-	-	-	-

⚠️ Safety / alignment disclaimer (important)

This is a preview experiment and has not been completely safety-aligned. The model may produce harmful, biased, or insensitive content, 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, including drafting, summarization, Q&A, and light reasoning.
Assessment of Norwegian writing style and quality.
Early evaluation of behaviour, language coverage (Norwegian, Bokmål, Nynorsk, and English), and quality.

Limitations

Preview quality; outputs may be unstable and may hallucinate.
Not safety-aligned; the model may follow harmful instructions or generate problematic content.
This release is an early SFT checkpoint, not a final production model.

Weights & formats

Transformers (original)

NbAiLab/borealis2-26b-a4b-preview (safetensors).

License

The license of this model is an adaptation of the Apache 2.0 license with additional use-based restrictions. In particular, users of the model are required to refrain from intentionally using the model to recreate data the model has been trained on. The license also requires users not to use the model or its output to provide end-user services whose primary purpose is to give access to licensed press publications in the training data.

For more information, see the LICENSE and the License FAQ.

Acknowledgements

Thanks to the Gemma team at Google and to everyone contributing feedback on this preview.

Disclaimer

These models may have bias and/or any other undesirable distortions. When third parties, deploy or provide systems and/or services to other parties using any of these models (or using systems based on these models) or become users of the models, they should note that it is their responsibility to mitigate the risks arising from their use and, in any event, to comply with applicable regulations, including regulations regarding the use of artificial intelligence. In no event shall the owner of the models (The National Library of Norway) be liable for any results arising from the use made by third parties of these models.