Transformers: Universal Sentence Encoder

[rdisipio]

🌟New model addition

Model description

Encoder of greater-than-word length text trained on a variety of data.

Open Source status

• [ ] the model implementation is available: see paper https://arxiv.org/abs/1803.11175

• [ ] the model weights are available: available from tfhub: https://tfhub.dev/google/universal-sentence-encoder/4

• [ ] who are the authors: Google

Additional context

Standard for sentence embedding, but would like to compare to other methods without having to rely on Tensorflow hub.

[adai183]

+1 !!
At reply.ai we have been using USE a lot for Semantic Retrieval. What most impressed us was the Q&A dual encoder model. Works better than anything else I know in case you need semantic similarity between a query and contexts.
It's true that Tensorflow Hub makes it super easy to work with. But we use your Transformers lib for everything else. So would be nice to have it all in one place.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[Peter-Devine]

Bump. I would appreciate this, as it would be handy to have a model geared toward semantic similarity rather than auto-encoding/ auto-regression, as all of the other default models are.
Thanks!

[setu4993]

+1. I think this could be a great addition.

[andbloch]

+1

[setu4993]

This might be a more appropriate model to port: https://tfhub.dev/google/universal-sentence-encoder-multilingual-large/3

The model in the OP was updated with the one linked above, alongwith addition of 15 languages.

[fonhorst]

+1.

[MiroFurtado]

+1 or https://ai.googleblog.com/2020/08/language-agnostic-bert-sentence.html

[setu4993]

I started working on this independently here. Would be great to get some help from anyone interested to get it done faster.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.