🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/179182

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

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:

• Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

✅ You can merge normally! (5 Unrelated Failures)

As of commit 7aff08c with merge base c44b2f2 ():

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

The committers listed above are authorized under a signed CLA.

• ✅ login: satishkc7 / name: Satish K C (7aff08c)

@claude review this PR

Claude finished @mikaylagawarecki's task in 1m 25s —— View job

Review of PR #179182

• Read the changed file and understand the diff
• Check existing patterns for multi-signature docstrings (e.g., torch.max, torch.min)
• Verify consistency across all four updated functions
• Verify against native_functions.yaml that out_dtype is indeed a separate overload
• Post review feedback

Summary

This PR correctly fixes the documented signatures for torch.addmm, torch.baddbmm, torch.bmm, and torch.mm. The previous docs showed out_dtype=None as an optional keyword-style parameter, but at the C++ dispatcher level these are separate overloads (e.g., bmm.dtype(Tensor self, Tensor mat2, ScalarType out_dtype)) where out_dtype is a required positional ScalarType, not an optional. Passing None at runtime raises a TypeError.

The fix follows the established pattern used by torch.max/torch.min — showing two .. function:: blocks with :noindex:, which is the right approach.

Detailed Feedback

Looks good overall. The changes are consistent across all four functions, the .. function:: / :noindex: pattern matches existing conventions, and out_dtype=None references are fully removed.

One minor observation:

Nit — the second overload blocks for addmm and baddbmm duplicate the beta, alpha, and out keyword args. This is technically correct for completeness (and torch.max similarly repeats its args in each signature block), so I'd consider this acceptable. Just noting that there's a small maintenance cost — if the description of beta/alpha changes in the future, it would need updating in two places per function. This is not a blocking concern.

LGTM — this is a clean, well-scoped docs fix.

@claude use the pr-review skill to review this

Claude finished @albanD's task in 2m 10s —— View job

PR Review: #179182

• Read the changed file and understand the diff
• Check existing patterns for multi-signature docstrings
• Verify against native_functions.yaml that out_dtype is separate overloads
• Check formatting consistency across all four functions
• Post review feedback

Summary

Clean docs fix that correctly addresses #172751. The previous signatures documented out_dtype=None as an optional kwarg, but in native_functions.yaml these are structurally separate overloads (e.g., bmm.dtype(Tensor self, Tensor mat2, ScalarType out_dtype)) where out_dtype is a required positional ScalarType. Passing None at runtime raises TypeError. The fix uses the established .. function:: / :noindex: pattern already used by torch.max, torch.min, torch.normal, etc.

Code Quality

The formatting is correct and consistent across all four functions:

• .. function:: at column 0 with :noindex: indented 3 spaces — matches existing convention
• out_dtype is correctly documented as a required positional arg (dtype) rather than optional
• The out_dtype description properly ends with a period (the old version was inconsistent — comma in some places, no punctuation in others)
• bmm and mm correctly use Keyword Args: (with capital A) matching their primary signature blocks, while addmm and baddbmm use Keyword args: (lowercase a) — also matching their respective primary blocks

One observation worth noting:

• addmm and baddbmm duplicate beta, alpha, and out kwargs in the second overload block. This is technically correct and matches how torch.max duplicates its args across overloads. However, bmm and mm avoid duplication by only listing the differing args plus {out}. Both approaches are acceptable per existing conventions, but the inconsistency between the four functions in this PR is slightly unfortunate. Since the addmm/baddbmm overloads have beta and alpha as keyword-only args that aren't in bmm/mm, duplicating them is reasonable for completeness — a reader of the out_dtype overload needs to know those kwargs exist.

Backward Compatibility

No BC concerns — this is a docs-only change with no functional code modifications.

Recommendation

Approve

This is a well-scoped, correct docs fix that follows established patterns. The changes accurately reflect the actual C++ dispatcher overload structure in native_functions.yaml.

@pytorchbot merge

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team