multiple on_train_epoch_start callbacks but only one on_train_epoch_end?

[zhh210]

I thought the number of on_train_epoch_start and on_train_epoch_end should be equal to the number of epochs. But when I passed the following callback function:

class MyPrintingCallback(Callback):

    def on_train_epoch_start(self, trainer, pl_module):
        print('Train epoch start for epoch: ', pl_module.current_epoch)
        
    def on_train_epoch_end(self, trainer, pl_module):
        # will run only once in the beginning
        print('Train step end for epoch: ', pl_module.current_epoch)

on_train_epoch_end is only called in the 0th epoch:

Training: -1it [00:00, ?it/s]Train epoch start for epoch: 0
Epoch 0: : 4875it [00:33, 143.57it/s, Train step end for epoch: 0
Train epoch start for epoch: 1
Epoch 1: : 0it [00:00, 7096.96it/s, loss=2.13e+09, v_num=37] Train epoch start for epoch: 2
Epoch 2: : 0it [00:00, 11335.96it/s, loss=2.13e+09, v_num=37]Train epoch start for epoch: 3
Epoch 3: : 0it [00:00, 12052.60it/s, loss=2.13e+09, v_num=37]Train epoch start for epoch: 4
Epoch 4: : 0it [00:00, 4301.85it/s, loss=2.13e+09, v_num=37]

Any idea why this is happening?

[zhh210]

It turns out that the issue is because I was chaining many data loaders with itertools.chain() which was called only once by lightning.