Title: Add MobileNetV2 pytest for Cortex-M backend

[psiddh]

Summary:

Added RemoveNoopPass (from ARM backend) to CortexMPassManager — it checks that
_clone_dim_order has matching input/output dtypes before removing it, so it won't strip any
actual type or dim_order conversions. This eliminates the identity clone from MV2's
flatten/view and unblocks the runtime.

Lowered qtol from 20 to 10 — passes reliably. Didn't go lower since calibration data is
random (torch.randn) so the numerical diff varies across runs. The argmax assertion provides
an additional correctness check independent of tolerance.

Removed xfail from test_dialect_mv2 (stable now), kept it on test_implementation_mv2 with
strict=False for potential flakiness in the serialization/runtime path.

Test plan

pytest -c backends/arm/test/pytest.ini backends/cortex_m/test/models/test_mobilenet_v2.py -v

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17300

• 📄 Preview Python docs built from this PR

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

❌ 10 New Failures, 1 Cancelled Job, 1 Unrelated Failure

As of commit 882c4c1 with merge base ba2516c ():

NEW FAILURES - The following jobs have failed:

• pull / test-moshi-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 3fff76b625fb4d30042b99abc387f80686cdaf10e910ff9d8a8da7db396a288c /exec failed with exit code 1
• pull / test-samsung-models-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 5b03df56700cc11a4ddb8401d6b94bed75500de314d2ab8d03990ecc28e767a6 /exec failed with exit code 127
• Test CUDA Builds / test-model-cuda-e2e (openai, whisper-large-v3-turbo, non-quantized) / linux-job (gh)
  RuntimeError: Command docker exec -t be16fd91333aa8bd0c089bbf15214810fc3bd35483d0f9b881e23788a944f9fe /exec failed with exit code 1
• Test CUDA Builds / test-model-cuda-e2e (openai, whisper-large-v3-turbo, quantized-int4-tile-packed) / linux-job (gh)
  RuntimeError: Command docker exec -t bb782a3809c52658a682e5bfbc566ac836e96cb1e4a687d673163639f0420273 /exec failed with exit code 1
• Test CUDA Builds / test-model-cuda-e2e (openai, whisper-large-v3-turbo, quantized-int4-weight-only) / linux-job (gh)
  RuntimeError: Command docker exec -t 767220b3b3f4254a09374888d3d60df22dae492316bbe5d10cac932946eefbc1 /exec failed with exit code 1
• Test CUDA Builds / test-model-cuda-e2e (openai, whisper-small, non-quantized) / linux-job (gh)
  RuntimeError: Command docker exec -t 31ba1c274e2cd2f4d8c7f498b00f62849db35178f804a75cf08e8ebd73db384c /exec failed with exit code 1
• Test CUDA Builds / test-model-cuda-e2e (openai, whisper-small, quantized-int4-tile-packed) / linux-job (gh)
  RuntimeError: Command docker exec -t 0bdefff9fe54ac2f48a00f007bdff7c53f074aafe69fb620397ba3d5f36ac961 /exec failed with exit code 1
• Test CUDA Builds / test-model-cuda-e2e (openai, whisper-small, quantized-int4-weight-only) / linux-job (gh)
  RuntimeError: Command docker exec -t 6eec09bbc69d996f173e27db9d17576edd62b77b3da4acf0db9eca4b211dc84e /exec failed with exit code 1
• Test Metal Backend / test-model-metal-e2e (openai, whisper-large-v3-turbo, non-quantized) / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1
• Test Metal Backend / test-model-metal-e2e (openai, whisper-small, non-quantized) / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

CANCELLED JOB - The following job was cancelled. Please retry:

• pull / test-samsung-quantmodels-linux / linux-job (gh)
  ##[error]The operation was canceled.

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• pull / test-models-linux (linear, xnnpack-quantization-delegation, linux.2xlarge) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

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

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[psiddh]

Note : test_implementation_mv2 is marked XFAIL because MV2's graph contains a dim_order_ops._clone_dim_order.default
op (from the flatten/view before the classifier) that MV3 doesn't have. The runtime doesn't include a kernel for
this op (use_portable_ops=False).

Should we add ARM's RemoveNoopPass to CortexMPassManager to remove identity clone ops.

[AdrianLundell]

Yes if it is a noop I think it makes sense to remove, just make sure that it does not do any dim_order/ type conversion first. Also see if you can lower the numerical tolerance compared to mv3 and still get it to pass, would be interesting to know.

[rascani]

Shouldn't this be on the test_implementation_mv2?

[psiddh]

Yes you're right, Removed xfail from test_dialect_mv2 (stable now), kept it on test_implementation_mv2 with
strict=False for potential flakiness in the serialization/runtime path.

[Copilot]

Pull request overview

Adds a new MobileNetV2 model test for the Cortex-M backend and updates the Cortex-M pass pipeline to remove certain no-op operations during lowering, aiming to unblock the MV2 runtime path.

Changes:

• Add a new test_mobilenet_v2.py pytest covering both dialect and implementation paths (with calibration + top-1 argmax sanity check).
• Add RemoveNoopPass to CortexMPassManager pass list to eliminate identity _clone_dim_order/related no-op nodes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
backends/cortex_m/test/models/test_mobilenet_v2.py	Introduces MobileNetV2 Cortex-M tests (dialect + implementation) with op-count assertions and calibration.
backends/cortex_m/passes/cortex_m_pass_manager.py	Inserts RemoveNoopPass into the Cortex-M lowering pass sequence to remove no-op ops earlier in the pipeline.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The xfail reason uses a placeholder ticket id ("MLETORCH-XXX"), which makes it hard to track and resolve the expected failure. Please replace it with the real issue/task identifier (or a concrete explanation if no tracker exists).

Suggested change

	"mobilenet_v2": "MLETORCH-XXX - Investigate mobilenet_v2 flakiness"
	"mobilenet_v2": "Flaky on Cortex-M due to nondeterministic quantized kernels; keep xfailed until kernels are made deterministic or flakiness is resolved (no external ticket)."

[Copilot]

Adding RemoveNoopPass here may be unsafe for Cortex-M graphs: the current implementation in backends/arm/_passes/remove_noop_pass.py removes dim_order_ops._clone_dim_order / _to_dim_order_copy based only on dtype, without checking whether dim_order (layout) actually changes. That can drop real dim-order conversions. It will also likely change existing Cortex-M op-count expectations (e.g., tests that currently expect _clone_dim_order to remain after transforms). Consider either (a) extending/wrapping the pass to verify dim_order equivalence via node meta (similar to backends/transforms/remove_clone_ops.py), or (b) limiting removal to proven-identity cases, and update affected op-count tests accordingly.