imagine 4x A6000 with 10min gen time it would be dope
imagine 4x A6000 with 10min gen time it would be dope
Fyep and cheap too
Ahow can we push the devs ? maybe you could implement it for us or not possible?
Fi am trying to make kaggle work haha
Fnope i cant
Akaggle?
AI can't imagine if 50min on H100 how much it will be on kaggle 
Ftrying 1.3b
JDoc i just heard Wan can also create audio
Fthey show
Fbut they didnt publish yet
Jok, so we still need MMAudio
Fcurrently yes we have it already also 
Ji know, but you know it also needs an update for the 5000er 
Fworking on fps increase thing
Fi can do now
Fgood reminding
A@Dr. Furkan Gözükara can I do good inference with a 700$ gpu?
Alike a 4070
Fit is good as long as fits into vram
Fyou can also get 2nd used rtx 3090
Fshould be even cheaper
Jah okay awesome, thanks! Will check!
Afrom Newegg?
Fdepends on where you live
AGreece
Fsadly this wont work yet
Fno torchaudio for windows for cuda128 
Fbut still testing
Fye no windows. remind me tomorrow
Fwow this is really good compare
FAfps boost ?
Fyes
Fworking on implementation
�@Dr. Furkan Gözükara Can the wan2.1 14b model run on 48 gb vram?
Ayes
Ait uses 43.9gb
Fyes it runs even at 24 gb
Fi didnt test 12gb yet but i expect work with sufficient RAM
F16 gb runs too
�I am not asking about your gradio app. I am asking about installing it using their instructions on github and using the unquantized models.
Ffull model uses over 50gb
Factually i am not sure how much full model uses
Fbut with cpu offloading i think should work
Ffine on 48 gb
�I ran the installation according to their instructions on github and got cuda out of memory error. I will try the cpu offload now but not able to get their gradio running
Fok it is getting ready
FFyou can ask on their github