🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Context matters and this is why late chunking is important As per usual, when something works great out-of-the-box, it indicates that you can squeeze some more gain by training for it!

On some tasks (where I essentially need to reformulate and restructure long documents I attach) GPT-4.5 is awesome, O3 hallucinates like crazy... It's gonna be sad to see it go, clearly no other model fills the gap

Reducing ColPali / ColQwen index size is super valuable in many use case, and I know many people who tried and couldn't beat the clustering technique from Ben Clavié. Clustering aware training helps ! Cool work Yubo Ma ! arxiv.org/abs/2506.04997

Amazing work on evals! Now, time for big techs to work on improving their VLMs for multi-page document understanding (instead of focusing on gaming Needle In A Haystack evals)