nb-sbert

[!NOTE] As of April 13th 2026, there are now new versions of this model, with improved performance and increased context length.

• NbAiLab/nb-sbert-v2-base
• NbAiLab/nb-sbert-v2-large

NB-SBERT-BASE is a SentenceTransformers model trained on a machine translated version of the MNLI dataset, starting from nb-bert-base.

The model maps sentences & paragraphs to a 768 dimensional dense vector space. This vector can be used for tasks like clustering and semantic search. Below we give some examples on how to use the model. The easiest way is to simply measure the cosine distance between two sentences. Sentences that are close to each other in meaning, will have a small cosine distance and a similarity close to 1. The model is trained in such a way that similar sentences in different languages should also be close to each other. Ideally, an English-Norwegian sentence pair should have high similarity.

Embeddings and Sentence Similarity (Sentence-Transformers)

As seen above, using the library sentence-transformers makes the use of these models quite convenient:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer, util
sentences = ["This is a Norwegian boy", "Dette er en norsk gutt"]

model = SentenceTransformer('NbAiLab/nb-sbert-base')
embeddings = model.encode(sentences)
print(embeddings)

# Compute cosine-similarities with sentence transformers
cosine_scores = util.cos_sim(embeddings[0],embeddings[1])
print(cosine_scores)

# Compute cosine-similarities with SciPy
from scipy import spatial
scipy_cosine_scores = 1 - spatial.distance.cosine(embeddings[0],embeddings[1])
print(scipy_cosine_scores)

# Both should give 0.8250 in the example above.

Embeddings and Sentence Similarity (HuggingFace Transformers)

Without sentence-transformers, you can still use the model. First, you pass in your input through the transformer model, then you have to apply the right pooling-operation on top of the contextualized word embeddings.

from transformers import AutoTokenizer, AutoModel
import torch

#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)

# Sentences we want sentence embeddings for
sentences = ["This is a Norwegian boy", "Dette er en norsk gutt"]

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('NbAiLab/nb-sbert-base')
model = AutoModel.from_pretrained('NbAiLab/nb-sbert-base')

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print(embeddings)

# Compute cosine-similarities with SciPy
from scipy import spatial
scipy_cosine_scores = 1 - spatial.distance.cosine(embeddings[0],embeddings[1])
print(scipy_cosine_scores)

# This should give 0.8250 in the example above.

SetFit - Few Shot Classification

SetFit is a method for using sentence-transformers to solve one of major problem that all NLP researchers are facing: Too few labeled training examples. The 'nb-sbert-base' can be plugged directly into the SetFit library. Please see this tutorial for how to use this technique.

Keyword Extraction

The model can be used for extracting keywords from text. The basic technique is to find the words that are most similar to the document. There are various frameworks for doing this. An easy way is to use KeyBERT. This example shows how this can be done.

pip install keybert

from keybert import KeyBERT
from sentence_transformers import SentenceTransformer
sentence_model = SentenceTransformer("NbAiLab/nb-sbert-base")
kw_model = KeyBERT(model=sentence_model)

doc = """
De første nasjonale bibliotek har sin opprinnelse i kongelige samlinger eller en annen framstående myndighet eller statsoverhode.
Et av de første planene for et nasjonalbibliotek i England ble fremmet av den walisiske matematikeren og mystikeren John Dee som
i 1556 presenterte en visjonær plan om et nasjonalt bibliotek for gamle bøker, manuskripter og opptegnelser for dronning Maria I
av England. Hans forslag ble ikke tatt til følge.
"""
kw_model.extract_keywords(doc, stop_words=None)

# [('nasjonalbibliotek', 0.5242), ('bibliotek', 0.4342), ('samlinger', 0.3334), ('statsoverhode', 0.33), ('manuskripter', 0.3061)]

The KeyBERT homepage provides other several interesting examples: combining KeyBERT with stop words, extracting longer phrases, or directly producing highlighted text.

Topic Modeling

To analyse a group of documents and determine the topics, has a lot of use cases. BERTopic combines the power of sentence transformers with c-TF-IDF to create clusters for easily interpretable topics.

It would take too much time to explain topic modeling here. Instead we recommend that you take a look at the link above, as well as the documentation. The main adaptation you would need to do to use the N