Fix torch.meshgrid Indexing Deprecation in PyTorch

When working with coordinate grids in PyTorch, you may see a deprecation warning related to the usage of torch.meshgrid. This warning appears because PyTorch is tightening the function's API and will require users to explicitly specify how grid indexing should be handled.

Historically, torch.meshgrid relied on an implicit indexing convention. While this behavior worked, it often led to confusion - especially for users coming from NumPy - since different libraries interpret grid dimensions differently. To avoid ambiguity and make code behavior explicit, PyTorch now asks developers to provide an indexing parameter.

The following example uses torch.meshgrid without specifying the indexing mode:

import torch

x = torch.tensor([0, 1])
y = torch.tensor([0, 1, 2])

xx, yy = torch.meshgrid(x, y)
print(xx) # tensor([[0, 0, 0], [1, 1, 1]])
print(yy) # tensor([[0, 1, 2], [0, 1, 2]])

Output:

UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument.

The deprecation warning can be fixed by adding indexing='ij', ensuring the output remains identical to the previous example:

import torch

x = torch.tensor([0, 1])
y = torch.tensor([0, 1, 2])

xx, yy = torch.meshgrid(x, y, indexing='ij')
print(xx) # tensor([[0, 0, 0], [1, 1, 1]])
print(yy) # tensor([[0, 1, 2], [0, 1, 2]])