MEDIUM 5.3

PYSEC-2025-202

Details

PyTorch before 3.7.0 has a bernoulli_p decompose function in decompositions.py even though it lacks full consistency with the eager CPU implementation, negatively affecting nn.Dropout1d, nn.Dropout2d, and nn.Dropout3d for fallback_random=True.

Are you affected?

Enter the version of the package you're using.

Affected packages

PyPI / torch

Introduced in: 2.6.0 Fixed in: 2.7.0

Fix pip install --upgrade 'torch>=2.7.0'

References

https://github.com/pytorch/pytorch/compare/v2.6.0...v2.7.0 [WEB]
https://gist.github.com/shaoyuyoung/4bcefba4004f8271e64b5185c95a248a [ADVISORY]
https://gist.github.com/shaoyuyoung/e636f2e7a306105b7e96809e2b85c28a [ADVISORY]
https://github.com/pytorch/pytorch/issues/142853 [REPORT]
https://github.com/pytorch/pytorch/pull/143460 [FIX]