Closed sunlibocs closed 3 years ago
FYI, I trained 1 epoch of this config except for the batch size is 8. It uses around 1 hour for an epoch on 4 of Tesla V100 GPU.
It should take around 40 minutes for one epoch using 8 GPUs, so @surfii3z's comment sounds about right. What is your configuration, @sunlibocs?
Training the packet will take 9 hours for one epoch by using the default setting. It will take ~20 days to train 50 epochs.
How much time did you spend training the model?
So sorry for my troubles.