# machine-learning
- h
haru0l
09/06/2021, 8:44 AMk80s does work - h
haru0l
09/06/2021, 8:45 AMthe only reason it isnt recommended is because it doesn't have a large amount of memory - k
kingximo
09/06/2021, 8:46 AMthen what do u suggest? internet explorer? lol - k
kingximo
09/06/2021, 8:49 AMyea but i cant even get p100 v100 or p4 - h
haru0l
09/06/2021, 8:54 AMif you have the money, either go buy a gpu, get colab pro, or just stick with a k80 - h
hecko
09/06/2021, 9:00 AMthose are pretty much reserved for paying users - h
hecko
09/06/2021, 9:01 AMgetting colab pro is probably the most cost-effective, unless you're in it for the really long haul and also somehow your gpu doesn't break after 5 years - k
kingximo
09/06/2021, 9:40 AMbreh - s
Scramble
09/06/2021, 2:32 PMyou can usually get like 1 lucky t4 as an non paying user. - s
Scramble
09/06/2021, 2:32 PMI've gotten 2 lucky t4's - h
haru0l
09/06/2021, 2:41 PMi just accepted it and train on k80s - s
Scramble
09/06/2021, 2:54 PMyeah - s
Scramble
09/06/2021, 2:54 PMit works - h
haru0l
09/06/2021, 2:55 PMtraining talknet on k80 kinda feels like the same amount of time to me but on a lower batch size - s
Scramble
09/06/2021, 2:57 PMluckily for training I got a lucky t4 - s
Scramble
09/06/2021, 2:59 PMdon't you love when the talknet model doesn't compile it's self so you have to manually compile the stuff and then have to wait for it upload (you probably don't have to but I had to) - m
mega b
09/06/2021, 3:16 PMadwhdawihaw - c
Cioss
09/06/2021, 4:23 PMwe should probably have non-character voices like a kid crowd yelling, movie announcer, kids toy ad or anything else no? could be used for youtube videos or satire commercials or whatever - z
Zephyr
09/06/2021, 5:04 PMSo uh, when is the graph gonna appear? - m
mega b
09/06/2021, 5:04 PMoh jeez - z
Zephyr
09/06/2021, 5:06 PMAny solutions? So far, the model I'm trying to train uses around 3:58 seconds of data. It's at epoch 14 and still hasn't shown a graph. - h
haru0l
09/06/2021, 5:28 PMgraph literally went 🚶 - h
haru0l
09/06/2021, 5:28 PMtry waiting a bit more, hopefully it'll at least show something - h
haru0l
09/06/2021, 5:29 PMotherwise, add more data - z
Zephyr
09/06/2021, 5:30 PMYeah. Seems like I'll probably need to add more data. - b
BIA
09/06/2021, 8:13 PMdoes anyone know why hifigan cpu usage is super high? are there less processor intensive inference models? - b
BIA
09/06/2021, 8:15 PMmy local machine has an i9-11900 and a single sentence shoots up the CPU to 50%