Does torch.unbind() create a view or a copy?

1

Does torch.unbind() create a tuple of views or copes of the original tensor i.e. is new memory allocated?

What about when applied to shared_memory tensors?

Is there a way to re-bind them?

2

Based on this code snippet it’s a view:

x = torch.randn(3, 10)
y = torch.unbind(x, 0)
y[0].zero_()
print(x)
# tensor([[ 0.0000e+00,  0.0000e+00,  0.0000e+00,  0.0000e+00,  0.0000e+00,
#           0.0000e+00,  0.0000e+00,  0.0000e+00,  0.0000e+00,  0.0000e+00],
#         [ 1.1271e+00, -1.6113e-01, -9.3410e-01,  1.2211e-01,  8.4843e-01,
#           4.7476e-01, -8.4909e-01, -1.1142e-01, -4.4881e-01, -3.7674e-01],
#         [-1.5271e+00, -1.0888e+00, -1.3062e-03, -2.3746e-01, -1.6353e+00,
#           1.0766e+00,  4.8345e-01, -6.0865e-01,  8.8547e-02,  3.6673e-01]])