Is it necessary to fine-tune the *BERT base models from HuggingFace before being able to use the pooled output to perform similarity search?

Support
,
1

Hi,

is it necessary to fine-tune the *BERT base models from HuggingFace before being able to use the pooled output to perform similarity search? Or would they work out-of-the box?

Cheers,
Riccardo

2

Hi! Thank you for your question. When you use the sentence-transformers models (sentence-transformers (Sentence Transformers)) they are already fine-tuned and can directly be used.

Other models, like bert-base-uncased, must be fine-tuned before they are useful.