There is the following issue on this page: https://docs.pytorch.org/docs/stable/generated/torch.unique.html

The documentation does not mention the important nuance when using the argument dim - since the output has to have equal sizes in the dimensions not specified by dim , we will get some repetitions, rather than a unique set of values, to force all the unique sets to be represented in equal length torch tensors.

Example:
```
>>> import torch
>>> x = torch.tensor([[1, 3, 2, 3], [1,2,1,2]], dtype=torch.long)
>>> torch.unique(x,dim=0)
tensor([[1, 2, 1, 2],
[1, 3, 2, 3]])
>>> torch.unique(x,dim=1)
tensor([[1, 2, 3],
[1, 1, 2]])
```

cc @svekars @sekyondaMeta @AlannaBurke