✨ Enhancement on the output presentation

[Autochthonal]

Hi there!

The model will print the progress of loading the Validation dataloader redundantly besides each training epoch:

Epoch 147: 100%|██████████| 782/782 [00:18<00:00, 41.75it/s, v_num=2, train_loss=3.360, Acc%=37.70]
Validation: |          | 0/? [00:00<?, ?it/s]
Validation:   0%|          | 0/157 [00:00<?, ?it/s]
Validation DataLoader 0:   0%|          | 0/157 [00:00<?, ?it/s]
Validation DataLoader 0:   1%|          | 1/157 [00:00<00:00, 217.64it/s]
Validation DataLoader 0:   1%|▏         | 2/157 [00:00<00:01, 124.62it/s]
Validation DataLoader 0:   2%|▏         | 3/157 [00:00<00:01, 100.94it/s]
Validation DataLoader 0:   3%|▎         | 4/157 [00:00<00:01, 95.37it/s] 
Validation DataLoader 0:   3%|▎         | 5/157 [00:00<00:01, 92.54it/s]
Validation DataLoader 0:   4%|▍         | 6/157 [00:00<00:01, 91.46it/s]
......
Validation DataLoader 0: 100%|██████████| 157/157 [00:01<00:00, 92.35it/s]
Epoch 148: 100%|██████████| 782/782 [00:17<00:00, 45.04it/s, v_num=2, train_loss=4.180, Acc%=38.80]

I wonder whether there exists a parameter to avoid it while presents the validation acc in each training epochs just like:

Epoch 147: 100%|██████████| 782/782 [00:18<00:00, 41.75it/s, v_num=2, train_loss=3.360, Acc%=37.70]
Epoch 148: 100%|██████████| 782/782 [00:17<00:00, 45.04it/s, v_num=2, train_loss=4.180, Acc%=38.80]
Epoch 149: 100%|██████████| 782/782 [00:21<00:00, 37.12it/s, v_num=2, train_loss=3.970, Acc%=36.50]

Thanks alot!

[o-laurent]

Hi @Autochthonal!

I've just tried a run and the validation progress bar has the same behavior as the training progress bar. It also disappears before switching to the next epoch, as shown here:

Sanity Checking: |                                                                                                                                          | 0/? [00:00<?, ?it/s]
Epoch 2: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 391/391 [00:16<00:00, 23.56it/s, v_num=104, train_loss=0.923, Acc%=66.20]`Trainer.fit` stopped: `max_epochs=3` reached.                                                                                                                                    
Epoch 2: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 391/391 [00:17<00:00, 22.71it/s, v_num=104, train_loss=0.923, Acc%=66.20]
Files already downloaded and verified
Files already downloaded and verified
Restoring states from the checkpoint path at logs/resnet18/standard/version_104/checkpoints/epoch=2-step=1173.ckpt
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1,2,3]
Loaded model weights from the checkpoint at logs/resnet18/standard/version_104/checkpoints/epoch=2-step=1173.ckpt
Testing DataLoader 0: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 79/79 [00:01<00:00, 45.49it/s]

What is the resolution of your screen? Sometimes, when the resolution is too small or when you change the window size while training, the progress bars are not updated correctly. @alafage, do you have an idea for this?

Have a nice day!

[alafage]

I agree with @o-laurent. I only have this weird behavior when I change the terminal window size while training. We cannot easily change that since we rely on Lightning for the progress bar. Let us know if you think there might be a bug on our side we missed.

Best!