Hadrien Reynaud

Thank you, was having the same issue !

@axel588 Your question is very vague and my answer will not help much. I am training on ~7k videos on 8xA100 and the videos start looking "okay" after 20 hours....

Hey @lucidrains, the v-parameterization is working well indeed and seems to converge faster (in actual time) than the ElucidatedImagen. But the sampling time is around 20 minutes for 250 steps...

@lucidrains The speed is not crucial, but I am reaching good scores on the task I want, and the sampling speed is still a significant drawback compared to previous methods...

Hi @lucidrains, how are those proteins doing? I was able to implement algorithm 2 of [Progressive Distillation...](https://arxiv.org/abs/2202.00512) and tried to make it work for v-parametrization. So far it's looking like...

Hi Phil, no worries, your work has helped me a tremendously already. I came across this paper this morning and it looks very promising indeed ! We'll definitely have 2-step...

For anyone who stumbles upon this thread, here is handy function to replace text at the paragraph level while keeping the format. It assumes that the format is the same...