fix zoedepth initialization error under deepspeed zero3

[Tavish9]

What does this PR do?

This PR addresses an issue with the initialization of tensor shapes in the ZoeDepth model when using the DeepSpeed framework with ZeRO-3 optimization.

Under DeepSpeed ZeRO-3, model parameters are partitioned, and the torch.Tensor class’s initialization behavior is overridden (details can be found in here). This override alters the behavior of tensor initialization, causing the error of the following line:

self.register_buffer("k_minus_1", torch.Tensor([self.k - 1]).view(1, -1, 1, 1), persistent=False)
print(self.k_minus_1.shape)  # torch.Size([1, 63, 1, 1]) wrong!

To resolve this, the initialization has been updated to use torch.tensor instead of torch.Tensor, which preserves the intended shape under DeepSpeed ZeRO-3.

self.register_buffer("k_minus_1", torch.tensor([self.k - 1]).view(1, -1, 1, 1), persistent=False)
print(self.k_minus_1.shape)  # torch.Size([1, 1, 1, 1]) correct!

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[qubvel]

Looks good to me, thanks!

cc @muellerzr