Fix: Conditionally import torch.distributed.algorithms.join in accelerator.py

[0xDELUXA]

What does this PR do?

torch.distributed is not available in all PyTorch builds (for example, Windows ROCm). Importing torch.distributed.algorithms.join.Join unconditionally at the top level of accelerator.py causes an immediate crash with:

ModuleNotFoundError: No module named 'torch._C._distributed_c10d'; 'torch._C' is not a package

This makes import accelerate fail entirely on those builds, even for users who have no intention of using distributed training.

Changes

• Wrapped from torch.distributed.algorithms.join import Join in if torch.distributed.is_available():
• Join is only used inside join_uneven_inputs() which is documented as multi-GPU DDP only, so this guard is safe and consistent with how the rest of the file handles optional distributed functionality

Prior art

The same fix was applied to huggingface/diffusers in:

• huggingface/diffusers#12420
• huggingface/diffusers#12425

A previous fix for the same class of issue in huggingface/transformers was made in:

• huggingface/transformers#40038

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• 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?

Who can review?

@SunMarc

[SunMarc]

Thx !

[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.

[0xDELUXA]

@SunMarc huggingface/transformers#44507 was recently merged. This one should be good to go as well. Thanks!