Hacker News new | ask | show | jobs
by lanceflt 755 days ago
They didn't train the vision encoder either, it's unchanged SigLIP by Google.
1 comments

“We finetuned billions of dollars of research by Google and Meta.”