How to repeat a tensor in order (with PyTorch)?

vision
1

If I have a 3 x 768 tensor [a, b, c], how can I repeat it such that I get [[a,b,c],[a,b,c],[a,b,c]], instead of [[a,a,a],[b,b,b],[c,c,c]], which is the case when I use torch.repeat_interleave? thanks!

2

I think you can do it this way:

n = 3 
m = 4 # 768 (I replace just to be able to visualize the result)

a = torch.tensor([1]*m)
b = torch.tensor([2]*m)
c = torch.tensor([3]*m)

abc = torch.stack([a, b, c], dim=0)

print(abc)

"""
tensor([[1, 1, 1, 1],
        [2, 2, 2, 2],
        [3, 3, 3, 3]])
"""

print(abc.repeat(3, 1, 1))
"""
tensor([[[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]],

        [[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]],

        [[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]]])
"""
# or
print(torch.repeat_interleave(abc.unsqueeze(0), repeats=3, dim=0))
"""
tensor([[[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]],

        [[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]],

        [[1, 1, 1, 1],
         [2, 2, 2, 2],
         [3, 3, 3, 3]]])
"""