reshape, broadcast_to, and permute_dims (I think these are all but I could have missed one), all accept a Tuple[int, ...]. The shape parameter for reshape and broadcast_to for shape, and axis for permute_dims

Every other function that accepts this kind of parameter accepts Union[int, Tuple[int, ...]].

It was pointed out in data-apis/array-api-compat#28 that NumPy currently accepts ints for all of these (and indeed, numpy.array_api accidentally does), but PyTorch does not.