Slice LazyTensor of shape (M,N) but with ndim=1

[tiangexiang]

Hi, I am trying to slice a lazy tensor in this way:

N, M, C = 5000, 2000, 10  
a = LazyTensor(torch.randn(1, N, C))  
b =  LazyTensor(torch.randn(M, 1, C))
c = (a-b).norm2() 
print(c.shape) # this gives [M, N]
d = c[10:20]
print(d.shape) # this gives an error, but I hope to get the shape [M, 10]

However, I encountered an error saying "Starting index is out of bounds" with ndim is actually 1.
How can I correctly slice the tensor as the way I expect?

Thank you!

[JonasDeSchouwer]

Just as a comment to this question, ndim currently doesn't work properly in PyKeops (see #376). So you should assume c.shape is correct.

[JonasDeSchouwer]

To actually solve your question (pbb a lot too late), can you try:
d = c[:, 10:20]