[JingyaHuang]

What does this PR do?

Rewrites the Neuron guide (docs/source/en/optimization/neuron.md) for native TorchNeuron support and gives it the same layout as the TPU guide.

AWS Trainium page

• Replaces the old Optimum Neuron export flow with the native torch_neuronx backend: pipe.to("neuron") and torch.compile(backend="neuron").
• Sections:
  • Eager mode (FLUX.2-klein-4B, single NeuronCore)
  • Compiled mode (static shapes, compile_repeated_blocks)
  • Tensor parallelism (Qwen-Image-2.1 with TensorParallelConfig; text encoder stays on CPU; fixed-seed CPU generator)
• Notes the NEFF compile cache and TORCH_NEURONX_NEFF_CACHE_DIR.
• Keeps a short pointer to Optimum Neuron for ahead-of-time exported pipelines.

Google TPU page

• Adds the optimized eager mode (EagerMode.DEFER_AND_FUSE, enabled with TORCH_TPU_DEFER_AND_FUSE=1) to the execution-mode table.

Both pages

• Pages are titled by hardware ("Google TPU", "AWS Trainium"). The intro explains that TorchTPU / TorchNeuron is the PyTorch backend that runs Diffusers on that hardware.
• Sidebar entries in _toctree.yml renamed to match.

Validation

All Neuron snippets were run unmodified on trn2 (Neuron SDK 2.33 / torch-neuronx 2.14.3), and each produced a correct 1024² image:

Snippet	Cores	Result
Eager, FLUX.2-klein-4B, 4 steps	1	✅
Compiled, FLUX.2-klein-4B, 4 steps	1	✅
Tensor parallel, Qwen-Image-2.1, 40 steps	8	✅
Tensor parallel + compile_repeated_blocks	8	✅

The TPU snippets weren't run; only the mode table changed there.

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?

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.

[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.

[stevhliu]

thanks for the docs! 🚀

[sayakpaul]

We should refer the readers to the dedicated TP section and show Neuron-relevant stuff here?

[sayakpaul]

Have we tested for this mode?

[JingyaHuang]

Yes we did, actually there are 3 eager modes... (there's a debug mode as well), I did not put in the beginning since I found it confusing and one should use torch.compile for deployment anyway, but indeed the semi-compiled mode is faster in between strict compile and compiled...

[sayakpaul]

Nice!