It is possible to extract the actual word embedding vectors when loading embeddings indexes from the Hugging Face Hub?

[namp]

I was following this tutorial

https://neuml.hashnode.dev/embeddings-in-the-cloud

and I'm not quite sure how to extract the actual word embedding vectors after this line of code:

embeddings = Embeddings() embeddings.load(provider="huggingface-hub", container="neuml/txtai-intro")

I need the actual word vectors for another task that I'd like to run in parallel.

Is it even possible?

Thanks

[davidmezzetti]

The embeddings index loaded from the Hugging Face Hub is the same format as other txtai indexes. The vectors are stored in Faiss. While it's possible to reconstruct the embeddings from Faiss, most of the time it's easier to call transform/batchtransform.

For example, the following snippet returns a vector per the embeddings settings.

embeddings.transform((None, "text", None))

or with a batch of text

embeddings.batchtransform([(None, "text1", None), (None, "text2", None)])

[davidmezzetti]

Closing due to inactivity. Re-open or open a new issue if there are further questions.