Presented ColPali at the ICLR Poster Session

I presented our poster “ColPali: Efficient Document Retrieval with Vision Language Models” at ICLR !

The original paper starts to be a bit dated as it was released in June, but I decided to create a poster that stays true to the original work. For all the novelties, follow me on X!

What about more recent stuff ?

Since the release, newer and better ColVision models were released. Most of them are on the online leaderboard but just to visualize a few:

Don’t forget about bi-encoder models, which work very well as well and can be more practical to deploy !

We also had to iterate on the ViDoRe benchmark as it was becoming too easy for recent models! While the V2 is a work in progress, and new tasks will get added, we managed to create harder tasks while keeping a consistent signal. Read more in the blogpost.

More cool things to come!