# machine-learning
- c
ColdFir3
09/22/2021, 1:55 PMYeahh you get 37hours of a p100 weekly - c
ColdFir3
09/22/2021, 1:56 PMYeahh you need a phone number, if u don't wanna use your personal number buy a sum card for like £1 and use that - c
ColdFir3
09/22/2021, 1:56 PMSomewhat easy just used the code from the Colab but adjusted it to fit for kaggle, took a while tho of day to night work after college, but no there's no tutorials - c
ColdFir3
09/22/2021, 1:56 PMI'll try to work on this soon 👀 - u
(Dawn) Will Draw Fictional Women
09/22/2021, 1:59 PMbtw idk how kaggle works but will there be some kind of release of this anytime soon - c
ColdFir3
09/22/2021, 2:01 PMPotentially I'll try to work on it a little more so it's more user friendly for people who don't know how to code, as it will require a fractional amount of CMD line knowledge - u
(Dawn) Will Draw Fictional Women
09/22/2021, 2:01 PMi see - c
Cris140
09/22/2021, 3:40 PMWindows has one - c
Cris140
09/22/2021, 7:33 PMDo you guys know how to load another hifi-gan model on colab? - c
Cris140
09/22/2021, 7:34 PMI tried inserting the ID in the colab but it gave me errors - h
hecko
09/22/2021, 7:36 PMplease post the errors - h
Hugh G. Rection
09/22/2021, 7:43 PMI think my Gromble needs retraining, because he sounds like a robot... - c
Cris140
09/22/2021, 8:12 PMI think I found the problem, my fault, I was trying to load a pretrained model instead of a model - c
Cris140
09/22/2021, 8:12 PMI was wanting to use another model because the universal one sounds pretty crap in some models - u
user
09/23/2021, 8:55 PMI really want to make decent quality models but there's 2 issues 1. I never get anything other than K80 2. I constantly run out of memory and it's annoying as hecc - u
user
09/23/2021, 8:56 PMI can't train my models because ooPS!11!!! you raN OUt of MEMory!!!!!!!!!!! - u
user
09/23/2021, 8:56 PMfun is temporary, pain is endless - m
mega b
09/23/2021, 9:01 PMK80 doesnt affect quality - m
mega b
09/23/2021, 9:01 PMrunning out of memory is fixable by lowering batch size - c
Couch
09/23/2021, 9:03 PMdont you have to change lr as well since it looks 50% less - j
JacobLenstar
09/23/2021, 9:04 PMwith talknet I have to set all batch sizes to 1 - m
mega b
09/23/2021, 9:04 PMinteresting - u
user
09/23/2021, 9:06 PMI know, but still Training is painfully slow also I didn't know you were supposed to lower batch size so thanks homie B) - c
Couch
09/23/2021, 9:06 PMmr cookie from the pony preservation project said something about learning rate and came up with a formula to find the lr based on batch size - c
Couch
09/23/2021, 9:06 PMits on github tt2 issues - m
mega b
09/23/2021, 9:06 PMyes - m
mega b
09/23/2021, 9:07 PMapplies to only Ljspeech & tacotron2 - m
mega b
09/23/2021, 9:08 PM0.001 X The square root of (batch size ÷ 256) - u
user
09/23/2021, 9:09 PMso mega b, if I have a k80, what's a batch size you'd recommend