Clarify multiple-inheritance op merge in TorchAOBaseTensor docstring

[PlatformLC]

Summary

Clarify a self-contradictory sentence in the TorchAOBaseTensor docstring about
op inheritance under multiple inheritance. Docstring-only; no code or behavior change.

Problem

Current text:

For multiple inheritance (e.g., class C(B, A)), ops are inherited from all parents
following Python's MRO (Method Resolution Order), with later bases taking priority.

The two clauses contradict each other: standard Python MRO gives the earlier
(leftmost) base priority, yet the sentence says the later base wins.

What the code actually does

__init_subclass__ walks cls.__bases__ in declaration order and merges each
parent's op table with dict.update
(see torchao/utils.py, the __init_subclass__ loop). A later-listed base
overwrites an earlier one on conflicts, so the later base does take priority —
but this is a flat overwrite merge, not standard MRO resolution. The original
sentence jumped from the MRO premise straight to the conclusion, skipping the
overwrite step that bridges them.

Change

Reword to state the mechanism directly, keeping MRO as the visit-order premise and
removing the contradiction:

For multiple inheritance (e.g., class C(B, A)), ops are inherited from all parents.
Each parent's table is merged in Python's MRO (Method Resolution Order), visiting B
then A; since each merge overwrites the previous, the later-listed base (A)
takes priority when both define the same op.

Notes

• torchao/utils.py only; no functional change, no test impact.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4955

• 📄 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.