[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/vt/ - Virtual Youtubers

Search:


View post   

>> No.75862532 [View]
File: 2.59 MB, 1264x1848, minato aqua_00038_.png [View same] [iqdb] [saucenao] [google]
75862532

>>75861818
>Default Outfit:
>minato aqua, minAto aqua \(1st costume\),
I typed in a single tag by hand and I made a typo fml

>>75861410
Money. A ton of money.
Averaging things out and spitting numbers without complete accuracy: Loras are trained for about 2000 steps and take about an hour on our GPUs. An entire model would need a few million images and multiple epochs of training on those images. Let's say, 2 million images and 50 epochs as a minimum. That is already 100 million steps. For 100 million steps on our own private hardware we would need 50000 hours to train the bare minimum model like pony. That time can be reduced if we distribute the workload on multiple GPUs that are more powerful than what we have and renting them costs about 2-3$/hr for one GPU. 5 GPUs would lower it let's say 20 times, So we're left with having to pay for 2500 hours of 2$/hr time 5 GPUs. At the very least, that's 25000$.

Navigation
View posts[+24][+48][+96]