Time required for training

[slives-lab]

I am wondering how much time is required to train a set of wav files. Currently, the status is: epoch: 1110.0000 (after running for 7 days using 1 GPU v100 50G RAM):

Here is the command that I used:

python jukebox/train.py --hps=small_vqvae --name=small_vqvae --sample_length=262144 --bs=4
--audio_files_dir=~/WAV/ --labels=False --train --aug_shift --aug_blend

[hitmusic100]

How many .wav files did you use; and what was the average duration of each? Did you succeed in your training and did you get results? Regards

[slives-lab]

I used 3,300 .wav files (12 sec each), which were split from longer files. I did not have success as it ran for >3 weeks but timed out due to exceeding my wall time allocation. Any advice?

[hitmusic100]

I wish - I would really like to train a model, since a lot of the artists in the v2_artist_ids.txt and v3_artist_ids.txt are obscure to me. The model openai has doesn't seem to have the very best (pop in my case) artists in the v2/v3 indexes. There must be a way !