docs/types(pt2e): annotate and document DuplicateDQPass (#4541)

[Anai-Guo]

Summary

Part of the PT2E type-hinting & docstring audit proposed in #4541. This PR
covers a single small, self-contained file:
torchao/quantization/pt2e/quantizer/duplicate_dq_pass.py.

• Remove the file-level # mypy: allow-untyped-defs.
• Add the one missing annotation: _maybe_duplicate_dq(...) -> None (it was the
  only untyped def; DuplicateDQPass.call and the nested maybe_replace_node
  were already annotated).
• Add module-, function-, class-, and method-level docstrings.

No behavior change — the executable lines are byte-for-byte identical; only the
mypy pragma, the one return annotation, and docstrings differ.

Why this is safe

allow-untyped-defs only suppresses checking of untyped function bodies, so
call was already type-checked and passing. The only newly-checked body is
_maybe_duplicate_dq, whose types are straightforward (GraphModule/Node
in, None out); the existing # type: ignore[assignment] on the
user.args/user.kwargs rewrites is preserved.

ruff check passes and no line exceeds 88 columns.

🤖 Generated with Claude Code

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4565

• 📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Unrelated MacOS MPS test jobs time out (30 min) due to stale cpp_extension lock file poisoning self-hosted runners

✅ No Failures

As of commit 53bc23b with merge base bfbc842 ():
💚 Looks good so far! There are no failures yet. 💚

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

[andrewor14]

@claude review please

[claude]

Claude finished @andrewor14's task in 37s —— View job

---

Code Review: DuplicateDQPass annotation & docs

• Read the PR diff and full file
• Verify no behavior change
• Check annotation & docstring accuracy

---

Verdict: LGTM ✅ — This does exactly what it claims: adds one return annotation, drops the file-level # mypy: allow-untyped-defs, and adds docstrings. No executable line changed.

Correctness

• No behavior change confirmed. The only functional lines touched are the signature of _maybe_duplicate_dq (annotation only) and the removed mypy pragma. The # type: ignore[assignment] on user.args/user.kwargs (lines 70–71) is preserved, which is correct — Node.args/kwargs are typed as immutable Argument tuples, so the assignments still need the ignore.
• -> None is accurate. Both return paths (return at line 58 and falling off the end) yield None.
• Docstrings match the code. The module/call docstrings correctly describe the choose_qparams -> getitem -> q -> dq dynamic-quant skip and the per-user cloning via annotation check. The _maybe_duplicate_dq docstring correctly notes unannotated users are left untouched.

Minor observations (non-blocking, not introduced by this PR)

• q_node = node.args[0] (line 92) is accessed as q_node.op without a Node type guard, unlike the nested getitem_node/choose_qparam_node checks which do use isinstance(..., torch.fx.node.Node). Since allow-untyped-defs is now removed for the module, it's worth confirming mypy is happy with call here — but this line is inside the already-type-checked call body and unchanged by this PR, so if CI is green it's a non-issue. Not something to fix in this docs-focused PR.
• Module docstring says "once per annotated user" while call's docstring says "cloned per user" — both are fine given the annotation gate lives inside _maybe_duplicate_dq; consistent enough.

Consistency with repo conventions

• Matches the pattern of the sibling PT2E type-hint/docstring PRs (part of Module-wise type hinting and docstring coverage audit for PT2E #4541). Line length stays ≤88 cols, RST-style :class: / double-backtick markup is consistent with the rest of the codebase.

Nice, tidy change. No requested changes from me.
• branch type-hints/pt2e-duplicate-dq-pass