Possible to get image embeddings out directly?

[nabsiddiqui]

Hello, would it be possible to get the image embeddings out directly? Say for each image in a folder?

[frankiethull]

As of now, this package will take images and encode them before returning an LLM's response. By embeddings, can you give an example of what the use case is or results you are after?

For the ollamar-backend, looks like there's a few ways to creating embeddings, we could possibly support.

For the ellmer-backend, I think there will be additional packages for embedding and RAG, so TBD on this side.

Let me know if you have an example and we can test support on the ollamar side, or if you are willing to put a PR, I can review it.

[nabsiddiqui]

I'm mainly interested in basic things, such as clustering of the images and cosine similarity between the vectors.

[frankiethull]

I guess embedding text is supported, but support for embedding images is TBD with ollama:
ollama/ollama#5304

will bump & track this issue to include image embedding!

[nabsiddiqui]

Thank you. No rush. I just noticed there was a need for image embeddings in R and this seemed like it was already doing something in that area.