Add device_name classmethod in Accelerator.

[GdoongMathew]

What does this PR do?

Fixes #17355

• 1 GPU

GPU available: NVIDIA GeForce RTX 3050 4GB Laptop GPU, using: 1 devices.
TPU available: False, using: 0 TPU cores
HPU available: False, using: 0 HPUs

• 2 GPUs (Kaggle)

>>> t = Trainer(devices=2)
INFO: GPU available: Tesla T4, using: 2 devices.
INFO: TPU available: False, using: 0 TPU cores
INFO: HPU available: False, using: 0 HPUs

• No cuda GPU available

GPU available: False, using: 0 devices.
TPU available: False, using: 0 TPU cores
HPU available: False, using: 0 HPUs

• MPS (Need Help)
• XLA (Kaggle)

GPU available: False, using: 0 devices.
TPU available: v3-8, using: 8 TPU cores
HPU available: False, using: 0 HPUs

Before submitting

• Was this discussed/agreed via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or minor internal changes/refactors)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing, make sure you have read the review guidelines. In short, see the following bullet-list:

Reviewer checklist

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

---

📚 Documentation preview 📚: https://pytorch-lightning--21112.org.readthedocs.build/en/21112/

[GdoongMathew]

Results in Kaggle:

• CPU: https://www.kaggle.com/code/gdoong/lightning-cpu?scriptVersionId=258513715&cellId=2
• GPU x 2: https://www.kaggle.com/code/gdoong/lightning-gpu?scriptVersionId=258514966&cellId=2
• TPU: https://www.kaggle.com/code/gdoong/lightning-tpu?scriptVersionId=258515194&cellId=2

[justusschock]

How would you handle heterogeneous gpu types? E.g. say I have one 3090 and one 4090 in my workstation. This doesn't handle it at all.

[GdoongMathew]

How would you handle heterogeneous gpu types? E.g. say I have one 3090 and one 4090 in my workstation. This doesn't handle it at all.

Hi @justusschock , based on the change, the output should look something like

GPU available: RTX 3090, RTX 4090, using: 2 devices.

that being said, if the setup is more complex - for example 2 x 3090 and 1 x 4090 - the current output might not fully reflect that.

[GdoongMathew]

On second thought, I may have oversimplified the question. It looks like the current implementation could cause issues with the DDP training strategy, since it tries to access a non-existent device ID on rank zero.

Thanks for your review @Borda & @justusschock , I’ll mark this PR as WIP for now. Any further suggestions are welcome~

[justusschock]

that being said, if the setup is more complex - for example 2 x 3090 and 1 x 4090 - the current output might not fully reflect that.

@GdoongMathew I think that's fine. it's just important to reflect all available gpu types as that might impact memory etc.

[GdoongMathew]

On second thought, I may have oversimplified the question. It looks like the current implementation could cause issues with the DDP training strategy, since it tries to access a non-existent device ID on rank zero.

Thanks for your review @Borda & @justusschock , I’ll mark this PR as WIP for now. Any further suggestions are welcome~

To follow up on my own concern: it seems the device_ids property refers to devices on the current node, not all devices across the world view. So the current implementation probably won’t cause any issues, aside from not being able to list device types from other nodes.