Make WanTransformer3DModel hidden states contiguous before the block loop

[menglcai]

What does this PR do?

Similar to #14186

The non-contiguous tensor originates from hidden_states.flatten(2).transpose(1, 2) produces a non-contiguous layout whose overhead accumulates across transformer blocks in WanTransformer3DModel.

This patch improves performance on both CUDA and ROCm, though the gain is more pronounced on ROCm.

Benchmarks (Wan-AI/Wan2.2-TI2V-5B-Diffuser, 1280x704, 50 steps, bf16):

Platform	PyTorch	Default	+contiguous	Speedup
NVIDIA RTX 5090	2.12.1+cu132	5.38 s/it	5.22 s/it	+3.0%
AMD Radeon AI PRO R9700 (gfx1201)	2.12.0+rocm7.15.0a20260713	12.23 s/it	9.62 s/it	+21.3%
AMD Radeon(TM) 8060S Graphics (gfx1151)	2.12.0+rocm7.15.0a20260711	115.73 s/it	79.30 s/it	+31.5%

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Did you read the Coding with AI agents guide?
  • Did you run the self-review skill on the diff?
  • Did you share the final self-review notes in the PR description or a comment?
• Did you read the contributor guideline?
• Did you read our philosophy doc? (important for complex PRs)
• 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?
• Are you the author (or part of the team) of the model/pipeline (only applicable for model/pipeline related PRs)?

Who can review?

@sayakpaul

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.

[github-actions]

Hi @menglcai, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. Fixes #1234) to the PR description so the issue is linked. See the contribution guide for more details. If this PR intentionally does not fix a tracked issue, a maintainer can add the no-issue-needed label to silence this reminder.

[sayakpaul]

So the pattern is that we identify the first occurrence of ops that can render the tensor in a non-contiguous layout and call contiguous() on it?

[menglcai]

Yes, the pattern is to ensure the inputs (especially hidden_states) are contiguous before entering the transformer block loop (see also #14186, #14203). We observed that ops like flatten+transpose or pos_embed can leave tensors in a non-contiguous memory layout, and calling .contiguous() before the loop can improve performance. The benefit is most significant on ROCm.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.