search

starrytong / SCNet

MIT License
59 stars 5 forks source link

About the training time & training details #16

Open LuckyHeart opened 2 months ago

LuckyHeart commented 2 months ago

Thank you for your great work.

I’m currently trying to train a model and would like to ask how long you trained using 8 Nvidia V100 GPUs. Could you please share some tips or important points to consider during the training process?

Thank you for your help.

starrytong commented 2 months ago

On the musdb18hq dataset, the large version requires 3 days of training and needs to be trained with AMP (another branch). The regular version only takes about 1 day. Perhaps you can start with the regular version first. I have updated the model code (added normalization), so you just need to run the train command (that's what I did).