Evaluation on STS and RTE tasks

[ruanchaves]

I'd just like to bring to your attention that the large portuguese-BERT model has been evaluated on the semantic textual similarity and textual entailment tasks in the ASSIN datasets.

Portuguese-BERT has achieved state-of-the-art Pearson correlation results on the similarity task, if compared to architectures that do not utilize automatic translation.

For more information please refer to this blog post or to my repository for the ASSIN datasets.