# machine-learning
- u
user
07/21/2021, 12:31 AMI don't I'm afraid. - a
Ask for dataset = $25/hour
07/21/2021, 12:31 AMI've got 10gb to work with in my 3080, but I just learned the notebooks recently, and havent tried to set it up locally yet - e
Eli.
07/21/2021, 12:31 AMSV2TTS used like ~4.5GB for me - e
Eli.
07/21/2021, 12:32 AMYeah Jupyter notebook is pretty nice - a
Ask for dataset = $25/hour
07/21/2021, 12:32 AMepoch 134 looking fresh - u
user
07/21/2021, 12:32 AM - e
Eli.
07/21/2021, 12:32 AM3080s are value if you chain them together when the price isnt insane anymore - a
Ask for dataset = $25/hour
07/21/2021, 12:33 AMi wish i could understand you because it seems important that I do - a
Ask for dataset = $25/hour
07/21/2021, 12:33 AMthis is 90 lines, so kinda modest? - u
user
07/21/2021, 12:34 AMAm pretty dumb, you don't need to lol. - e
Eli.
07/21/2021, 12:34 AMWhich model is this? - a
Ask for dataset = $25/hour
07/21/2021, 12:34 AM89 actually, i had to scrap one 30 second clip - e
Eli.
07/21/2021, 12:34 AMAs in architecture my bad - a
Ask for dataset = $25/hour
07/21/2021, 12:35 AMI'm using the uberduck notebook to process Michael Garfield from the Future Fossils podcast, but i'm sure thats not quite the answer you expected - a
Ask for dataset = $25/hour
07/21/2021, 12:35 AMnew to all of this - e
Eli.
07/21/2021, 12:35 AMUu - e
Eli.
07/21/2021, 12:35 AMThat's tacotran2 right? - a
Ask for dataset = $25/hour
07/21/2021, 12:35 AMyes - e
Eli.
07/21/2021, 12:35 AMYeah no problem - u
user
07/21/2021, 12:35 AMThe gif is in a good way. Wanted the face more than the "oof", it was more of an oooooh. - a
Ask for dataset = $25/hour
07/21/2021, 12:35 AMthanks bud - e
Eli.
07/21/2021, 12:35 AMWanted to ask about overfitting + transfer learning with that model, but no worries if you're not super experienced with it yet - a
Ask for dataset = $25/hour
07/21/2021, 12:36 AMi know 500 epochs can cause overfitting - e
Eli.
07/21/2021, 12:36 AMI'd imagine depends on the dataset size? - a
Ask for dataset = $25/hour
07/21/2021, 12:36 AMsurely - u
user
07/21/2021, 12:36 AMI kinda get the overfitting idea but can you explain again? Is it just the case that you won't get good results for new words? - e
Eli.
07/21/2021, 12:37 AMOverfitting is where the model matched the data you train it on too closes- rather than learn the overall 'pattern', it learns how to reproduce your exact data - e
Eli.
07/21/2021, 12:37 AMSo when you utilize it on a real use case, it wont be able to work. - a
Ask for dataset = $25/hour
07/21/2021, 12:37 AMmaking a TTS that is only good for the words you put in to it sounds like a colossal disappointment - u
user
07/21/2021, 12:38 AMCool yea that makes sense, thanks