Too many slices, some is empty

When we use the pretrained resnet and sequeezenet, the training is end with this problem.
I don’t know which cause this problem.

@qingyangxu Are you running training on multiple GPUs ? This error can come if multiple GPUs are enabled in training but the training size is not multiple of number of GPUs.

Can you try changing dataset split on the data tab in a way where

((training size) % batch_size) / NUM_GPUs = 0

1 Like

Yes, we are training on two gpus.
I will try changing dataset split.

Thank you very much.