UKPLab / sentence-transformers

State-of-the-Art Text Embeddings
https://www.sbert.net
Apache License 2.0
15.17k stars 2.47k forks source link

Is 'stsb-roberta-base' already fine-tuned or just a structure? #735

Closed happygirlzt closed 3 years ago

happygirlzt commented 3 years ago

Dear there, many thanks for your amazing work.

According to the models listed on README, I have trouble understanding 'stsb-roberta-base'. Is this already fine-tuned on STSbenchmark? Now, I have my own dataset, duplicate and non-duplicate sentence pairs from another domain. I wonder whether it works if I further fine-tune on this model (i.e., 'stsb-roberta-base'). Or I build my own Siamese BERT structure to fine-tune the Siamese BERT on my own data. Many thanks in advance!

nreimers commented 3 years ago

It is fine-tuned on NLI and STSb. You can use it as start for your training