fix: add PyTorch version check for torch.accelerator in gather_size_by_comm

[Mr-Neutr0n]

Summary

Add backward compatibility check for torch.accelerator.current_accelerator() which only exists in PyTorch 2.6+.

Problem

gather_size_by_comm() in _modeling_parallel.py uses torch.accelerator.current_accelerator() without checking if it exists. Since diffusers officially supports PyTorch 2.1+, this causes AttributeError: module 'torch' has no attribute 'accelerator' on PyTorch versions 2.1-2.5.

Solution

Added hasattr(torch, "accelerator") check with fallback to "cuda" for older PyTorch versions. This matches the pattern already used in other parts of the codebase:

• src/diffusers/loaders/lora_pipeline.py (line 111)
• src/diffusers/hooks/group_offloading.py (lines 121-125)
• src/diffusers/quantizers/gguf/gguf_quantizer.py (lines 159-163)

Fixes#13074

[Mr-Neutr0n]

Friendly bump! Let me know if there's anything I should update or improve to help move this forward.

[DN6]

I think this can actually be reduced to a one liner if we use

diffusers/src/diffusers/utils/torch_utils.py

Line 281 in 2843b3d

defget_device():

gather_device="cpu"if"cpu"incomm_backendselseget_device()

[Mr-Neutr0n]

@DN6 — done. Pushed as 3a12e93 on the same branch: the three-branch if/elif/else is now a one-liner that uses get_device() from torch_utils. Behavior is preserved (cpu for cpu backends, otherwise the appropriate device via get_device(), which already handles accelerator version compatibility internally).

[qgallouedec]

This contribution comes from a profile that opens a massive number of pull requests generated purely by AI; I'm closing it.