val dataloader batch size overwrites train dataloader batch size

[sebastiangonsal]

I am using two different dataloaders for train and validation. These two dataloaders use two different datasets.
Mysteriously, the batch size for val_dataloader (which is defined after train_dataloader) overwrites the batch size for train_dataloader. As a result, the training code uses batch size that is batch size for validation. Is this a known bug?

This is how it is defined:

    train_loader = DataLoader(train_dataset, batch_size=64, \
        collate_fn=get_collate_fn())

    val_loader = DataLoader(val_dataset, batch_size=1, \
        collate_fn=get_collate_fn())

And then passed to pytorch lightning's trainer.fit:

   trainer.fit(model, train_loader, val_loader)