GPL for sentence embedding tasks?

[hanshupe]

In the provided examples GPL us used for semantic search tasks: given a query, relevant results should be retrieved. Is it also the recommended approach to get meaningful embeddings / bi-encoders, or is it better to use TSDAE?

[rbbby]

I was wondering the same thing, any update on this?
Intuitively it should work, but it is unclear how it compares to solely using TSDAE or using a both methods.

[artmatsak]

I'm curious about this, too. We'd like to do domain adaptation for all-mpnet-base-v2.

[gabriead]

Any update on this?

[rbbby]

I went for TSDAE with finetuning on nli and sts corpora (there are scripts on the sentence transformer github that you can use). Seems to have worked great!