Hi,
this is very weird.
I have two tensors, inds and scores
where inds.ndim=1 and scores.ndim=1
and inds.shape=(28), scores.shape=(300)
the content is:
inds tensor([ 1, 2, 18, 3, 0, 11, 12, 16, 14, 163, 41, 104, 87, 139,
176, 74, 304, 143, 261, 265, 191, 257, 316, 106, 267, 241, 118, 253],
device='cuda:0')
when calling
a=scores[inds]
pytorch would throw
IndexError: too many indices for tensor of dimension 1
but calling
torch.gather(scores, 0, inds) is fine
any idea?