What is the rationale behind transposing the input to fastai’s DataLoader? In particular why do we pass transpose=True in the IMDb notebook (trn_dl = DataLoader(trn_ds, bs//2, transpose=True, num_workers=1, pad_idx=1, sampler=trn_samp)), but stick to the default transpose=False in most other cases.
Also, does the standard torch.utils.data.DataLoader’s collate_fn parameter perform a similar function?