Hi HN, I’m one of the two authors of the post and the Linum v2 text-to-video model (https://news.ycombinator.com/item?id=46721488). We're releasing our Image-Video VAE (open weights) and a deep dive on how we built it. Happy to answer questions about the work!
Hadn’t seen that before! Seems very in line with what with the broader points about regularization. In table 4 they show faster convergence in 200 epochs when used alongside REPA. I’d be curious to see if it ended up beating REPA by itself with full 800 epochs of training — or if something about this new latent space, leads to plateauing itself (learns faster but caps out on expressivity). We’ve seen that phenomena before in other situations (eg UNET learns faster than DiT because of convolutions, but stops learning beyond a certain point).
This seems like a great model to experiment fine tuning with original art, given it’s relatively small and with open license. Is that a fair assessment?
Thanks for the great write up and making it available to us all.
The kind that I like so much on HN. It tickle your mind but is still clear enough for an advanced beginner.
Thanks for the great write up and making it available to us all.