# machine-learning
- c
Couch
11/17/2021, 9:15 PMit only outputs files after hparams.epochs have been trained - u
user
11/17/2021, 9:18 PMso why is it outputting numbers sofast - c
Couch
11/17/2021, 9:19 PMno clue - c
Couch
11/17/2021, 9:19 PMare you getting gpu activity - u
user
11/17/2021, 9:19 PMyes 2000 mb vram usage - c
Couch
11/17/2021, 9:19 PMOH WAIT - u
user
11/17/2021, 9:19 PM???? - c
Couch
11/17/2021, 9:19 PMmaybe the reason why its going so fast is that youre using a very large batch size - u
user
11/17/2021, 9:19 PMno its 64 - u
user
11/17/2021, 9:20 PMthe hparams is default - c
Couch
11/17/2021, 9:20 PMthe number of iterations per epoch is# of lines in list.txt/batch size - c
Couch
11/17/2021, 9:21 PMmaybe it is possible that your dataset is too small for the batch size ? - u
user
11/17/2021, 9:21 PMit has 39 - u
user
11/17/2021, 9:21 PMis 64 too much for 39 - u
user
11/17/2021, 9:21 PMoh of course - c
Couch
11/17/2021, 9:22 PMwell i guess we found the problem - c
Couch
11/17/2021, 9:22 PMalso i highly suggest you set
for a small dataset like thatlearning_rate=1e-5 - c
Couch
11/17/2021, 9:22 PMyou can use the colab's batch size which is 14 (
)batch_size=14 - u
user
11/17/2021, 9:23 PMok good idea - u
user
11/17/2021, 9:23 PMnew problem - u
user
11/17/2021, 9:23 PMit can not understand the wav? - u
user
11/17/2021, 9:24 PMdo you have an example list.txt and wavs you can send to me? - u
user
11/17/2021, 9:24 PMwith small dataset - u
user
11/17/2021, 9:24 PMmaybe my wav is wrong - c
Couch
11/17/2021, 9:24 PMyour list.txt is correct - u
user
11/17/2021, 9:24 PMbut it worked on a colab - c
Couch
11/17/2021, 9:24 PMits just that in the colab the wavs are converted to mel spectrograms (which is numpy data, so file extension
)npy - u
user
11/17/2021, 9:24 PMoh - u
user
11/17/2021, 9:24 PMso this is an expected problem?