# ISL-SBERT small

Sentence transformer, trained using unsupervised technique, TSDAE. The models take in an Icelandic text and creates sentence embeddings from it.

Based off of this [article](https://www.pinecone.io/learn/unsupervised-training-sentence-transformers/)


## Data

the model was trained on 100_000 sentences selected at random from clarin-is: [unanotated news2 from IGC(RMH)](https://repository.clarin.is/repository/xmlui/handle/20.500.12537/238)

to install the data, run the following command:

```bash
curl --remote-name-all https://repository.clarin.is/repository/xmlui/bitstream/handle/20.500.12537/238{/IGC-News2-22.10.TEI.zip}
```

## Author

Sigurður Haukur Birgisson