Fix PyTorch 2.3.1 compatibility: add version guard for torch.library.… by Aishwarya0811 · Pull Request #12206 · huggingface/diffusers

Aishwarya0811 · 2025-08-21T09:11:23Z

What does this PR do?

Fixes #12195 by adding version guards for torch.library.custom_op and torch.library.register_fake which are not available in PyTorch 2.3.1.

Problem

torch.library.custom_op and torch.library.register_fake were introduced in PyTorch 2.4
Users with PyTorch 2.3.1 get AttributeError: module 'torch.library' has no attribute 'custom_op'
This breaks from diffusers import AutoencoderKL and other imports

Solution

Added hasattr() checks before using these torch.library functions
Functions are only registered when available in the PyTorch version
Maintains backward compatibility with PyTorch 2.3.1 without breaking newer versions

Testing

✅ Tested with PyTorch 2.3.1 - import now works without errors
✅ Verified AutoencoderKL imports and functions correctly
✅ No regressions expected with newer PyTorch versions

Fixes #12195

Who can review?

@sayakpaul @yiyixuxu - This is a PyTorch compatibility fix for core library functionality.

…custom_op - Add hasattr() check for torch.library.custom_op and register_fake - These functions were added in PyTorch 2.4, causing import failures in 2.3.1 - Both decorators and functions are now properly guarded with version checks - Maintains backward compatibility while preserving functionality Fixes huggingface#12195

a-r-r-o-w

Hi, thanks for the fix! I think we can improve a bit here by using the approach followed in #11941. Specifically, the following lines:

diffusers/src/diffusers/models/attention_dispatch.py

Lines 121 to 139 in 0018b62

    
           if torch.__version__ >= "2.4.0": 
        
               _custom_op = torch.library.custom_op 
        
               _register_fake = torch.library.register_fake 
        
           else: 
        
               def _custom_op_no_op(name, fn=None, /, *, mutates_args, device_types=None, schema=None): 
        
                   def wrap(func): 
        
                       return func 
        
                   return wrap if fn is None else fn 
        
               def _register_fake_no_op(op, fn=None, /, *, lib=None, _stacklevel=1): 
        
                   def wrap(func): 
        
                       return func 
        
                   return wrap if fn is None else fn 
        
               _custom_op = _custom_op_no_op 
        
               _register_fake = _register_fake_no_op

- Replace hasattr check with version string comparison - Add no-op decorator functions for PyTorch < 2.4.0 - Follows pattern from huggingface#11941 as suggested by reviewer - Maintains cleaner code structure without indentation changes

Aishwarya0811 · 2025-08-21T12:29:24Z

Hi @a-r-r-o-w
Thanks for the feedback! I've updated the PR to use the dummy decorator approach as suggested, following the pattern from #11941. The code is much cleaner now without the indentation changes.

Tested with PyTorch 2.3.1 and the import works correctly.

a-r-r-o-w

Thanks for the updates! Just one more comment

src/diffusers/models/attention_dispatch.py

Update all the decorator usages Co-authored-by: Aryan <[email protected]>

Co-authored-by: Aryan <[email protected]>

Aishwarya0811 · 2025-08-22T20:26:19Z

Hi @a-r-r-o-w I've addressed all your feedback - moved the version check to the top of the file and used private naming with underscores.

a-r-r-o-w

Thanks for looking into it @Aishwarya0811!

a-r-r-o-w · 2025-08-23T07:02:05Z

@bot /style

github-actions · 2025-08-23T07:02:35Z

Style bot fixed some files and pushed the changes.

HuggingFaceDocBuilderDev · 2025-08-23T07:10:59Z

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.

#12206) * Fix PyTorch 2.3.1 compatibility: add version guard for torch.library.custom_op - Add hasattr() check for torch.library.custom_op and register_fake - These functions were added in PyTorch 2.4, causing import failures in 2.3.1 - Both decorators and functions are now properly guarded with version checks - Maintains backward compatibility while preserving functionality Fixes #12195 * Use dummy decorators approach for PyTorch version compatibility - Replace hasattr check with version string comparison - Add no-op decorator functions for PyTorch < 2.4.0 - Follows pattern from #11941 as suggested by reviewer - Maintains cleaner code structure without indentation changes * Update src/diffusers/models/attention_dispatch.py Update all the decorator usages Co-authored-by: Aryan <[email protected]> * Update src/diffusers/models/attention_dispatch.py Co-authored-by: Aryan <[email protected]> * Update src/diffusers/models/attention_dispatch.py Co-authored-by: Aryan <[email protected]> * Update src/diffusers/models/attention_dispatch.py Co-authored-by: Aryan <[email protected]> * Move version check to top of file and use private naming as requested * Apply style fixes --------- Co-authored-by: Aryan <[email protected]> Co-authored-by: Aryan <[email protected]> Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

sayakpaul requested a review from a-r-r-o-w August 21, 2025 09:54

a-r-r-o-w reviewed Aug 21, 2025

View reviewed changes

Aishwarya0811 requested a review from a-r-r-o-w August 21, 2025 12:42

a-r-r-o-w reviewed Aug 22, 2025

View reviewed changes

Aishwarya0811 and others added 6 commits August 23, 2025 00:24

Update src/diffusers/models/attention_dispatch.py

22dc9d1

Update all the decorator usages Co-authored-by: Aryan <[email protected]>

Update src/diffusers/models/attention_dispatch.py

7009fd8

Co-authored-by: Aryan <[email protected]>

Update src/diffusers/models/attention_dispatch.py

a831012

Co-authored-by: Aryan <[email protected]>

Update src/diffusers/models/attention_dispatch.py

6f799b0

Co-authored-by: Aryan <[email protected]>

Merge branch 'main' into fix-pytorch-231-compatibility

f515990

Move version check to top of file and use private naming as requested

5ef7da4

Aishwarya0811 requested a review from a-r-r-o-w August 22, 2025 20:27

a-r-r-o-w approved these changes Aug 23, 2025

View reviewed changes

Merge branch 'main' into fix-pytorch-231-compatibility

0a3a228

Apply style fixes

ec47936

a-r-r-o-w merged commit 9a7ae77 into huggingface:main Aug 23, 2025
11 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix PyTorch 2.3.1 compatibility: add version guard for torch.library.…#12206

Fix PyTorch 2.3.1 compatibility: add version guard for torch.library.…#12206
a-r-r-o-w merged 10 commits intohuggingface:mainfrom
Aishwarya0811:fix-pytorch-231-compatibility

Aishwarya0811 commented Aug 21, 2025 •

edited

Loading

Uh oh!

a-r-r-o-w left a comment

Uh oh!

Aishwarya0811 commented Aug 21, 2025

Uh oh!

a-r-r-o-w left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Aishwarya0811 commented Aug 22, 2025

Uh oh!

a-r-r-o-w left a comment

Uh oh!

a-r-r-o-w commented Aug 23, 2025

Uh oh!

github-actions bot commented Aug 23, 2025 •

edited

Loading

Uh oh!

HuggingFaceDocBuilderDev commented Aug 23, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	if torch.__version__ >= "2.4.0":
	_custom_op = torch.library.custom_op
	_register_fake = torch.library.register_fake
	else:

	def _custom_op_no_op(name, fn=None, /, *, mutates_args, device_types=None, schema=None):
	def wrap(func):
	return func

	return wrap if fn is None else fn

	def _register_fake_no_op(op, fn=None, /, *, lib=None, _stacklevel=1):
	def wrap(func):
	return func

	return wrap if fn is None else fn

	_custom_op = _custom_op_no_op
	_register_fake = _register_fake_no_op

Conversation

Aishwarya0811 commented Aug 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Problem

Solution

Testing

Who can review?

Uh oh!

a-r-r-o-w left a comment

Choose a reason for hiding this comment

Uh oh!

Aishwarya0811 commented Aug 21, 2025

Uh oh!

a-r-r-o-w left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Aishwarya0811 commented Aug 22, 2025

Uh oh!

a-r-r-o-w left a comment

Choose a reason for hiding this comment

Uh oh!

a-r-r-o-w commented Aug 23, 2025

Uh oh!

github-actions bot commented Aug 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

HuggingFaceDocBuilderDev commented Aug 23, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Aishwarya0811 commented Aug 21, 2025 •

edited

Loading

github-actions bot commented Aug 23, 2025 •

edited

Loading