Add OpenAI embedding feature

[pepeto]

It would be great to have this alternative in the combo.

Thank you! I Love Orage!!!

[PrimozGodec]

Hi, thank you for the suggestion. If I understand correctly, are these embeddings only available through their API? Is there no option to run models locally/on private infrastructure?

[pepeto]

It is correct, OpenAI Ada model is run via API (vector of 1536 elements). Nevertheless, there are other free options (also via API) in Hugging Face. In these case, the Model is downloaded just once when it's called for the first time. There are more info in this link and I think it could be very interesting to have these solutions at hand. Thank you!
https://huggingface.co/blog/getting-started-with-embeddings

[PrimozGodec]

Thank you for the suggestion. SBERT embedding in the paraphrase-multilingual-MiniLM-L12-v2 model that you can find at the huggingface.co. Would some other embedding (from those freely available) suit you better?

And while I was digging into this one, I just realised we didn't document what embedding we are using. We should improve the documentation.

[pepeto]

If you are connecting huggingface, there are many other models and you could even choose, select or enter the name of the model you like. All of them have the same interface in order to use them, and you don't know what can be added in the future. Take a look at Langchain and its possibilities of integrations. THANK YOU!!!

[PrimozGodec]

Huggingface API looks really promising. We should think about including it in Orange3-text.