EuropeanParliament

/

EUBERT

like 6

Hi,
Thanks for your question, I forgot to mention it, it's fixed.
For the moment, it's under [EUPL](License: eupl-1.1)

Can I ask you for more information about your usage and see if I can help you?
Best regards,
Sébastien

Thank you for adding licence information to this project.
I have tested your model.
I like that distribution of European languages are better than in CommonCrawl, where for example Slovak language has only 0.35% of all texts.
But, Europarl corpus it self is not suitable to train proper language properties and that is why this model was performing much worse than others testes on classification tasks.

Thanks for your feedback.
May I ask you if it's a classification task for Slovak documents only or for multilanguage documents ?
and which models are performing well for your use case ?

Hi,
yes it is classification is the Slovak language.
Those Slovak single language models are performing well.
TUKE-DeutscheTelekom/skroberta
gerulata/slovakbert

It makes sense that performance is better with a dedicated model.
Thanks again