Fix MaskedLayerNorm with bias=False

[adenzhou1350]

MaskedLayerNorm inherits nn.LayerNorm's bias=False option, but its forward always adds self.bias. With affine weights enabled and bias disabled, that is None and forward raises a TypeError.

Apply the weight as before and only add bias when present. This leaves the existing treatment of zero-valued inputs and default affine behavior unchanged.

Tests cover bias-free forward/backward against nn.LayerNorm on nonzero inputs (float32/float64, one- and two-dimensional normalized shapes), mixed/all-zero inputs, and existing affine options.

Validation on Linux, Python 3.12 and PyTorch 2.11.0: normal TorchAO source-package import with C++ extensions disabled and CUDA hidden; python3 -m pytest test/prototype/pat/test_masked_layernorm.py -q passed (4 tests, 7 subtests). The same tests fail on main with the Tensor + None error. Ruff check/format and diff check pass. No GPU or full-repository CI claim.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4966

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.