How the images / sequence diagrams / UML diagrams in the specifications are processed?

#2
by kspviswaphd - opened

First of all Kudos to your paper and thank you very much for opensourcing this dataset.

I have question on the processing of non-text data on the specs. As you know, 3GPP specs often come with UML flows such as sequence diagrams and NRM specs with inheritance relationships / ER diagrams. I wonder how these are processed as part of dataset and how these info are fed to LLM. For eg, I see the processed MD files contains links to the image files but I don't see them as part of dataset.

Screenshot 2024-08-12 at 11.44.25 PM.png

Thanks :)

The way we have converted the files preserves the image locations and files. We need to do some quality assurance to ensure the images and graphs are captured correctly, so we can update the dataset with those images. This could potentially be very useful for training multi-modal LLMs in the wireless domain.
Stay tuned !!

Sign up or log in to comment