You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello, I would like to ask about the training process, the batchsize is set to 1 to train 500epoch, I think the training time is a bit long, can I increase the batchsize to improve the training speed, but does this have an impact on the training results? Thank you!
The text was updated successfully, but these errors were encountered:
can I increase the batchsize to improve the training speed, but does this have an impact on the training results?
Based on my experience with CycleGAN-like models, increasing the batch size does not improve the training speed or the quality of the translations in most cases. In other words, generally one needs to train for 1M batches regardless of the batch size to get the same translation quality. However, the impact of the batch size can vary depending on your specific dataset. So, it might be beneficial to try increasing the batch size for your dataset to see if it helps.
Hello, I would like to ask about the training process, the batchsize is set to 1 to train 500epoch, I think the training time is a bit long, can I increase the batchsize to improve the training speed, but does this have an impact on the training results? Thank you!
The text was updated successfully, but these errors were encountered: