Add op test for torch.unique_consecutive

[Chase Naples (cnaples79)]

Summary

• Adds OpInfo-based test for torch.unique_consecutive operator
• Addresses issue [torchlib] Add op test to torch.unique_consecutive #2695

Changes

• Added sample_inputs_unique_consecutive function in extra_opinfo.py that filters samples from common_methods_invocations.sample_inputs_unique to only include cases with dim=None (as the implementation has limited dim support)
• Added OpInfo entry for ops.aten.unique_consecutive with integral_types dtypes (matching the implementation which only supports int32/int64)
• Added TorchLibOpInfo entry in ops_test_data.py to enable the test

Testing

The test follows the same pattern as other unique operator tests (_unique, _unique2, unique_dim) and reuses PyTorch's standard test infrastructure.

Fixes #2695

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.09%. Comparing base (20a99d1) to head (853c43a).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2742   +/-   ##
=======================================
  Coverage   70.09%   70.09%           
=======================================
  Files         226      226           
  Lines       27388    27388           
  Branches     2781     2781           
=======================================
  Hits        19198    19198           
  Misses       7234     7234           
  Partials      956      956           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Justin Chu (justinchuby)]

Thanks - looks like there are some errors

[Copilot]

Pull request overview

This PR adds OpInfo-based test coverage for the torch.unique_consecutive operator, addressing issue #2695. The test reuses PyTorch's standard test infrastructure and follows the established pattern used by other unique operators in the codebase.

Key changes:

• Added sample_inputs_unique_consecutive function to filter test samples to only include dim=None cases (matching implementation limitations)
• Added OpInfo entry for ops.aten.unique_consecutive with integral dtypes configuration
• Registered the operator in TorchLibOpInfo to enable the test

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
tests/function_libs/torch_lib/extra_opinfo.py	Adds sample input generator and OpInfo configuration for unique_consecutive operator
tests/function_libs/torch_lib/ops_test_data.py	Registers the operator in TorchLibOpInfo test suite

[Justin Chu (justinchuby)]

Thanks — but this OpInfo currently exposes a genuine lowering bug rather than just adding coverage, which is why the matrix is red. For return_inverse=True with multi-dim or scalar inputs, torch.unique_consecutive returns inverse shaped like the ORIGINAL input, but aten_unique_consecutive (~core.py L10031) computes inverse on the FLATTENED x and never reshapes it back; scalar () inputs (reshaped to (1,)) also mismatch torch's () inverse. Two ways forward: (a) fix aten_unique_consecutive to reshape inverse back to the input shape and handle rank-0, then keep the broad sampler; or (b) narrow the sampler to rank-1 inputs / add an xfail for return_inverse on multi-dim+scalar. Either gets CI green.

[krajani21]

Since PR #2742 has been inactive, I'd like to pick this up and implement the fix suggested by Justin Chu (@justinchuby).