embeddings instead of text output

[pasha76]

what is the best way to extract embeddings as output instead of text...We are planning to fine tune the model and extract embeddings insetad of text caption. Any suggestions?

[vikhyat]

Are you looking for image embeddings, or text embeddings?

[7AtAri]

I used the output_hidden_states=True, then you can access the text embeddings using output.hidden_states. You get the image embeddings by calling the vision encoder.

At least this is what I figured out.