[EASY] Always log 1st batch when resuming training

[bigning]

What does this PR do?

composer prints the console logging for 1st batch and every self.log_interval batches. but when resuming training, we probably want to print the 1st batch no matter if cur_batch % self.log_interval == 0. This PR just forces to print console logging for 1st batch after resumption.

test

1. firstly save checkpoint after training 3 batches.
2. load checkpoint and set console_log_interval=7ba composer train/train.py train/yamls/pretrain/mpt-125m.yaml train_loader.dataset.split=train_small max_duration=10ba eval_interval=0 console_log_interval=7ba load_path=./125m/checkpoints/ep0-ba3-rank0.pt

before the change:

first logged batch is 7

after the change

first logged batch is 4, then 7

[mvpatel2000]

Should we remove cur_batch==1 in this case?

Can we please add the same for def epoch_end?

[bigning]

Should we remove cur_batch==1 in this case?

Can we please add the same for def epoch_end?

good catch! updated.

[mvpatel2000]

LGTM! Thanks for the fix.