Fix SincShallowNet MPS backward failure

[Copilot]

SincShallowNet could fail during backward on Apple MPS because BatchNorm2d received a non-contiguous tensor from an einops axis permutation. CPU/CUDA were unaffected, but MPS BatchNorm backward can attempt an incompatible view on the saved input.

• Model change
  • Added an einops-based contiguous rearrange wrapper.
  • Used it before the depthwise BatchNorm2d so the saved BatchNorm input is contiguous.

class _ContiguousRearrange(Rearrange):
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return super().forward(x).contiguous()

• Regression coverage
  • Added a focused SincShallowNet test that verifies the depthwise BatchNorm receives contiguous input and that backward completes.

[Copilot]

Pull request overview

This PR addresses an Apple MPS-specific backward failure in SincShallowNet by ensuring the tensor passed into the depthwise BatchNorm2d is contiguous after an axis permutation.

Changes:

• Move the axis permutation out of the depthwiseconv nn.Sequential and into forward, applying .contiguous() after permute.
• Add a unit test that asserts the depthwise BatchNorm2d receives a contiguous input and that backward completes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
braindecode/models/sinc_shallow.py	Ensures contiguous memory layout before depthwise BatchNorm by permuting and calling .contiguous() in forward.
test/unit_tests/models/test_sinc_shallow.py	Adds a regression test that hooks the depthwise BatchNorm to assert its input is contiguous and that .backward() runs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ce4cb0e6bc

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve depthwiseconv checkpoint keys

For checkpoints saved before this change, the depthwise batch norm and conv weights are named depthwiseconv.1.* and depthwiseconv.2.*; starting this Sequential with BatchNorm2d renames them to depthwiseconv.0.* and depthwiseconv.1.*. That makes existing SincShallowNet checkpoints fail under default load_state_dict strict loading, and can silently leave these layers randomly initialized when users load non-strictly. Keeping a no-parameter contiguous rearrange wrapper at index 0, or adding a mapping, would avoid breaking saved weights.

Useful? React with 👍 / 👎.