Pyramid-Flow/pyramid-flow · Congratulations...finally!

Wow... I can't believe it - this is the first open source text 2 video model that is actually useful, like sora or runway gen 3. Do you happen to know how much vram is needed to serve the model and allow for 10 seconds / 24 fps videos?

And do you know how fast this model would run on an actual GPU (not a ZeroGPU?) Like an rtx3090 or 4090? I'm seriously considering renting a gpu rig from vast AI and using it to host this model... but only if it's going to be reasonably performant. I also wonder how fast it would run on a high end gpu, because if the speed is good enough I'll just rent an a100 and make it pay for itself...