Avoid fallback on CPU if no devices are provided

[rohitgr7]

What does this PR do?

Addresses: #12160 (comment)

Does your PR introduce any breaking changes? If yes, please list them.

Before submitting

• Was this discussed/approved 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 internal minor changes/refactorings)

PR review

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

• 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

Did you have fun?

Make sure you had fun coding 🙃

cc @Borda @justusschock @kaushikb11 @awaelchli @ninginthecloud @akihironitta @rohitgr7

[kaushikb11]

Could you update the docs as well?
For example: https://pytorch-lightning.readthedocs.io/en/latest/accelerators/gpu.html#select-gpu-devices

[DuYicong515]

Hi @rohitgr7, it makes sense to me Trainer(accelerator="gpu", devices=[]/0/"0") throws instead of falling back to cpu.

However, currently Trainer(gpus=[]/0/"0") still falls back to CPU.
Trainer(accelerator="gpu", gpus=[]/0/"0") will work as "auto" and use all GPU devices when GPUs are available , and throws if GPU is not available.

Even gpus is being deprecated as a Trainer argument, shall we still make the behaviour consistent here? I feel it's confused that those arguments behaves differently on similar settings.

[awaelchli]

However, currently Trainer(gpus=[]/0/"0") still falls back to CPU.

That's ok, it is documented here.

Trainer(accelerator="gpu", gpus=[]/0/"0") will work as "auto" and use all GPU devices when GPUs are available

Agreed, this is probably not intended and should be changed.

[kaushikb11]

@rohitgr7

Here, we assumed the accelerator is not available. What if it's available and user pass devices="0"/0/ []]?

It leads to such errors

    self._parallel_devices = self.accelerator.get_parallel_devices(self._devices_flag)
  File "/home/jovyan/pytorch-lightning/pytorch_lightning/accelerators/gpu.py", line 82, in get_parallel_devices
    return [torch.device("cuda", i) for i in devices]
TypeError: 'NoneType' object is not iterable

It is addressed in #12633