fix: add PyTorch version check for torch.accelerator in gather_size_by_comm#13093
Open
Mr-Neutr0n wants to merge 2 commits into
Open
fix: add PyTorch version check for torch.accelerator in gather_size_by_comm#13093Mr-Neutr0n wants to merge 2 commits into
Mr-Neutr0n wants to merge 2 commits into
Conversation
…y_comm torch.accelerator.current_accelerator() only exists in PyTorch 2.6+. Since diffusers officially supports PyTorch 2.1+, this causes AttributeError on versions 2.1-2.5. Added hasattr check with fallback to 'cuda' for older PyTorch versions, matching the pattern already used in lora_pipeline.py. Fixes huggingface#13074
Author
|
Friendly bump! Let me know if there's anything I should update or improve to help move this forward. |
DN6
reviewed
Feb 13, 2026
| elif hasattr(torch, "accelerator"): | ||
| gather_device = torch.accelerator.current_accelerator() | ||
| else: | ||
| gather_device = "cuda" |
Collaborator
There was a problem hiding this comment.
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
gather_device = "cpu" if "cpu" in comm_backends else get_device()
@DN6 (2026-02-13) suggested using the existing get_device() helper from torch_utils instead of the inline hasattr(torch, "accelerator") check. get_device() already returns the appropriate device string for the current environment and falls back to cpu when no accelerator is available, which makes the three-branch if/elif/else a one-liner.
Author
|
@DN6 — done. Pushed as |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
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.pyusestorch.accelerator.current_accelerator()without checking if it exists. Since diffusers officially supports PyTorch 2.1+, this causesAttributeError: 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