PyTorch before 3.7.0 has a bernoullip 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 fallbackrandom=True.
"https://github.com/pypa/advisory-database/blob/main/vulns/torch/PYSEC-2025-202.yaml"