Why are these logs displayed awkwardly in Jupyter Notebook

[JivanRoquet]

When running trainer.fit(model) in a Jupyter Notebook, the logs are printed the following way:

Epoch 0: 100%|█████████▉| 2315/2318 [00:31<00:00, 73.49it/s, loss=1.67, v_num=63, acc=0.625, lr=0.0005]
Validating: 0it [00:00, ?it/s]
Validating:   0%|          | 0/1 [00:00<?, ?it/s]
Epoch 0: 100%|██████████| 2318/2318 [00:35<00:00, 65.54it/s, loss=1.68, v_num=63, acc=0.472, lr=0.0005, val_loss=1.490, val_acc=0.580]
Epoch 1: 100%|█████████▉| 2315/2318 [00:31<00:00, 73.05it/s, loss=1.48, v_num=63, acc=0.594, lr=0.0004, val_loss=1.490, val_acc=0.580]
Validating: 0it [00:00, ?it/s]
Validating:   0%|          | 0/1 [00:00<?, ?it/s]
Epoch 1: 100%|██████████| 2318/2318 [00:36<00:00, 64.32it/s, loss=1.47, v_num=63, acc=0.667, lr=0.0004, val_loss=1.400, val_acc=0.599]
Epoch 2: 100%|█████████▉| 2315/2318 [00:32<00:00, 72.00it/s, loss=1.36, v_num=63, acc=0.578, lr=0.00032, val_loss=1.400, val_acc=0.599]
Validating: 0it [00:00, ?it/s]
Validating:   0%|          | 0/1 [00:00<?, ?it/s]
Epoch 2: 100%|██████████| 2318/2318 [00:35<00:00, 64.62it/s, loss=1.38, v_num=63, acc=0.583, lr=0.00032, val_loss=1.380, val_acc=0.606]
Epoch 0:   0%|          | 0/2318 [18:45:27<?, ?it/s]71it/s, loss=1.32, v_num=63, acc=0.594, lr=0.000256, val_loss=1.380, val_acc=0.606]
Epoch 3: 100%|█████████▉| 2315/2318 [00:30<00:00, 75.46it/s, loss=1.35, v_num=63, acc=0.648, lr=0.000256, val_loss=1.380, val_acc=0.606]
Validating: 0it [00:00, ?it/s]
Validating:   0%|          | 0/1 [00:00<?, ?it/s]
Epoch 3: 100%|██████████| 2318/2318 [00:34<00:00, 67.35it/s, loss=1.32, v_num=63, acc=0.722, lr=0.000256, val_loss=1.350, val_acc=0.611]
Epoch 4: 100%|█████████▉| 2315/2318 [00:30<00:00, 75.60it/s, loss=1.26, v_num=63, acc=0.680, lr=0.000205, val_loss=1.350, val_acc=0.611]
Validating: 0it [00:00, ?it/s]
Validating:   0%|          | 0/1 [00:00<?, ?it/s]
Epoch 4: 100%|██████████| 2318/2318 [00:34<00:00, 67.59it/s, loss=1.28, v_num=63, acc=0.500, lr=0.000205, val_loss=1.330, val_acc=0.616]

Based on the same logs, I'm guessing the "proper" way they should be displayed would be more like this:

Epoch 0: 100%|██████████| 2318/2318 [00:35<00:00, 65.54it/s, loss=1.68, v_num=63, acc=0.472, lr=0.0005, val_loss=1.490, val_acc=0.580]
Epoch 1: 100%|██████████| 2318/2318 [00:36<00:00, 64.32it/s, loss=1.47, v_num=63, acc=0.667, lr=0.0004, val_loss=1.400, val_acc=0.599]
Epoch 2: 100%|██████████| 2318/2318 [00:35<00:00, 64.62it/s, loss=1.38, v_num=63, acc=0.583, lr=0.00032, val_loss=1.380, val_acc=0.606]
Epoch 3: 100%|██████████| 2318/2318 [00:34<00:00, 67.35it/s, loss=1.32, v_num=63, acc=0.722, lr=0.000256, val_loss=1.350, val_acc=0.611]
Epoch 4: 100%|██████████| 2318/2318 [00:34<00:00, 67.59it/s, loss=1.28, v_num=63, acc=0.500, lr=0.000205, val_loss=1.330, val_acc=0.616]

The training/validation steps from the model I'm using are declared like the following:

    def training_step(self, train_batch, batch_idx):
        x, y = train_batch
        y_hat = self.forward(x)
        loss = cce(torch.log(torch.maximum(torch.tensor(1e-8).type_as(y_hat), y_hat)), y.argmax(dim=1))
        acc = tm.functional.accuracy(y_hat.argmax(dim=1), y.argmax(dim=1))
        self.log("loss", loss)
        self.log("acc", acc, prog_bar=True)
        self.log("lr", self.scheduler.get_last_lr()[-1], prog_bar=True)
        return loss

    def validation_step(self, valid_batch, batch_idx):
        x, y = valid_batch
        y_hat = self.forward(x)
        loss = cce(torch.log(torch.maximum(torch.tensor(1e-8).type_as(y_hat), y_hat)), y.argmax(dim=1))
        acc = tm.functional.accuracy(y_hat.argmax(dim=1), y.argmax(dim=1))
        self.log("val_loss", loss, prog_bar=True)
        self.log("val_acc", acc, prog_bar=True)

I've tried different combinations of the code above, but could not get rid of the log duplications when an epoch starts/ends.

Is there anything I'm doing wrong here? How to get to an output which is closer to the "proper" version shown above?

[rohitgr7]

working fine for me.. although tqdm can sometimes cause rendering issues. I'd suggest trying https://pytorch-lightning.readthedocs.io/en/latest/extensions/generated/pytorch_lightning.callbacks.RichProgressBar.html#richprogressbar for better rendering.