feature(device): add MUSAAccelerator to support Musa device

[chenkoucer]

What does this PR do?

Adds support for MUSA device by implementing MUSAAccelerator in PyTorch Lightning. This enables users to train models on MUSA accelerators with the same Lightning API.

Breaking Changes

• Support the Musa device

Before submitting

• Was this discussed/agreed via a GitHub issue?
• Did you read the contributor guideline?
• Did you make sure your PR does only one thing?
• Did you update the documentation?
• Did you write new necessary tests?
• Did you verify tests pass locally?
• Did you list all breaking changes?
• Did you update the CHANGELOG?

PR review

Anyone in the community is welcome to review the PR.

Reviewer checklist

• Is this pull request ready for review?
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory

---

📚 Documentation preview 📚: https://pytorch-lightning--21381.org.readthedocs.build/en/21381/

[bhimrazy]

Hi @chenkoucer,
Thanks for opening this PR and taking the time to work on this!
We really appreciate the contribution.

Since this is a significant new feature that introduces a new accelerator backend, it would be helpful to first open an issue describing the motivation, expected API usage, and any relevant references. For features of this scale, maintainers typically prefer an initial design discussion over the issue to make sure the approach and long-term maintenance implications are aligned. Without that prior context, it may be difficult for the team to evaluate or move the PR forward.

Also, I noticed that the implementation relies on torch.musa. I’m not sure whether this namespace exists in upstream PyTorch—could you also provide a reference to the official documentation or library that exposes this API? That would help reviewers verify the integration and understand the expected behavior.

Thanks again!

cc: @ethanwharris @justusschock