Notebooks

[smellslikeml]

Congrats to the team for the excellent work!
Just wanted to check if you can reupload the notebooks as the files appear to be empty.

[matthieu-perso]

Hi @smellslikeml, thanks for the interest and the kind words ! We've updated the notebooks.

[smellslikeml]

Much appreciated, @matthieu-perso

Your study has me thinking about applying similar techniques to understand how VLMs interpret 3D scenes.

So I'm trying out the idea of regressing distances between two objects in a scene on the embedding of the last token in distance estimation queries.
I'm comparing a model adapted to the task against the base model and seeing a lift in correlation between predicted and actual.

For the interested: https://colab.research.google.com/drive/17XqJdI8pd2B3dbqFQ1l25tkADNCyZ7dg?usp=sharing