Cased Finnish Sentence BERT model

Finnish Sentence BERT trained from FinBERT. A demo on retrieving the most similar sentences from a dataset of 400 million sentences can be found here.


  • Library: sentence-transformers
  • FinBERT model: TurkuNLP/bert-base-finnish-cased-v1
  • Data: The data provided here, including the Finnish Paraphrase Corpus and the automatically collected paraphrase candidates (500K positive and 5M negative)
  • Pooling: mean pooling
  • Task: Binary prediction, whether two sentences are paraphrases or not. Note: the labels 3 and 4 are considered paraphrases, and labels 1 and 2 non-paraphrases. Details on labels


The same as in the HuggingFace documentation of the English Sentence Transformer. Either through SentenceTransformer or HuggingFace Transformers


from sentence_transformers import SentenceTransformer
sentences = ["Tämä on esimerkkilause.", "Tämä on toinen lause."]

model = SentenceTransformer('TurkuNLP/sbert-cased-finnish-paraphrase')
embeddings = model.encode(sentences)

HuggingFace Transformers

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 = ["Tämä on esimerkkilause.", "Tämä on toinen lause."]

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('TurkuNLP/sbert-cased-finnish-paraphrase')
model = AutoModel.from_pretrained('TurkuNLP/sbert-cased-finnish-paraphrase')

# 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.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")

Evaluation Results

A publication detailing the evaluation results is currently being drafted.

Full Model Architecture

  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})

Citing & Authors

While the publication is being drafted, please cite this page.


  • J. Kanerva, F. Ginter, LH. Chang, I. Rastas, V. Skantsi, J. Kilpeläinen, HM. Kupari, J. Saarni, M. Sevón, and O. Tarkka. Finnish Paraphrase Corpus. In NoDaLiDa 2021, 2021.
  • N. Reimers and I. Gurevych. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. In EMNLP-IJCNLP, pages 3982–3992, 2019.
  • A. Virtanen, J. Kanerva, R. Ilo, J. Luoma, J. Luotolahti, T. Salakoski, F. Ginter, and S. Pyysalo. Multilingual is not enough: BERT for Finnish. arXiv preprint arXiv:1912.07076, 2019.
Downloads last month
Hosted inference API
Sentence Similarity
This model can be loaded on the Inference API on-demand.