support argmax/argmin without dim kwargs and fix adaptive_max_pool3d #14710

cccclai · 2025-09-30T22:03:52Z

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case

edit:

Apply to argmin too
Add exir_ops.edge.aten.adaptive_max_pool3d.default to the to be implemented op list to pass the error

Differential Revision: D83606497

pytorch-bot · 2025-09-30T22:03:55Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14710

📄 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:

An error occurred (AccessDenied) when calling the ... operation

❌ 1 New Failure, 23 Pending

As of commit 3a5bff8 with merge base 0e74a17 ():

NEW FAILURE - The following job has failed:

pull / test-samsung-models-linux / linux-job (gh)
RuntimeError: Command docker exec -t 0b21f27399435765559f180f9f2c2354d837f6809ea8664b1e0cf1fa4eaa8bcc /exec failed with exit code 1

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

facebook-github-bot · 2025-09-30T22:04:05Z

@cccclai has exported this pull request. If you are a Meta employee, you can view the originating Diff in D83606497.

github-actions · 2025-09-30T22:04:39Z

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.

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

facebook-github-bot · 2025-09-30T22:05:30Z

@cccclai has exported this pull request. If you are a Meta employee, you can view the originating Diff in D83606497.

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

facebook-github-bot · 2025-09-30T22:08:37Z

@cccclai has exported this pull request. If you are a Meta employee, you can view the originating Diff in D83606497.

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

facebook-github-bot · 2025-09-30T22:14:01Z

@cccclai has exported this pull request. If you are a Meta employee, you can view the originating Diff in D83606497.

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

facebook-github-bot · 2025-09-30T22:17:16Z

@cccclai has exported this pull request. If you are a Meta employee, you can view the originating Diff in D83606497.

haowhsu-quic · 2025-10-02T01:59:54Z

backends/qualcomm/_passes/insert_reshape_for_argmax.py

+from executorch.exir.pass_base import ExportPass, PassResult
+from executorch.exir.passes import dead_code_elimination_pass
+
+class InsertReshapeForArgmax(ExportPass):


I think we might have other ops like argmin which can leverage this pass as well. Do you mind rewording the class name into more generic one? Something like InsertReshapeForReduceOp.

I think the new commit wasn't pushed successfully. Could you try again? Thanks

oh sorry, just push

haowhsu-quic · 2025-10-02T02:00:01Z

backends/qualcomm/_passes/insert_reshape_for_argmax.py

+        graph = graph_module.graph
+        modified = False
+
+        for n in list(graph.nodes):


I think we can use the iterator directly: for n in graph.nodes:.

backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py

haowhsu-quic · 2025-10-02T06:02:51Z

backends/qualcomm/tests/test_passes.py

+import torch
+from executorch.backends.qualcomm._passes import InsertReshapeForArgmax
+
+class TestPasses(unittest.TestCase):


Thank you for adding this file, it's helpful for us to test all the passes thoroughly.

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

backends/qualcomm/builders/op_argmax.py

backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

cccclai · 2025-10-07T16:47:44Z

Can I get another round of review on this? I fix a few more failing tests including argmin, and adaptive_max_pool3d

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

meta-codesync · 2025-10-07T16:59:15Z

@cccclai has exported this pull request. If you are a Meta employee, you can view the originating Diff in D83606497.

backends/qualcomm/tests/test_qnn_delegate.py

Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

haowhsu-quic

Thank you!

kimishpatel · 2025-10-07T17:37:38Z

backends/qualcomm/partition/common_defs.py

 to_be_implemented_operator = [
    exir_ops.edge.aten._adaptive_avg_pool3d.default,
    exir_ops.edge.aten.adaptive_max_pool2d.default,
+    exir_ops.edge.aten.adaptive_max_pool3d.default,


is this pr just about argmax? i see maxpool added here. if it is about reduce ops then please do update the title

Yeah updated

…ytorch#14710) Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case Differential Revision: D83606497

cccclai · 2025-10-07T22:09:33Z

@pytorchbot cherry-pick --onto release/1.0 -c regression

…14710) Summary: As title, in PyTorch, when dim is not set, it will flatten the input and get argmax as dim=0. Add a pass to reshape the input when dim is not set and consolidate test case edit: 1. Apply to argmin too 2. Add `exir_ops.edge.aten.adaptive_max_pool3d.default` to the to be implemented op list to pass the error Differential Revision: D83606497 (cherry picked from commit e09abea)

pytorchbot · 2025-10-07T22:11:48Z

Cherry picking #14710

The cherry pick PR is at #14868 and it is recommended to link a regression cherry pick PR with an issue. The following tracker issues are updated:

[v1.0.0] Release Tracker #14288 (comment)

Details for Dev Infra team

Raised by workflow job

meta-cla bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Sep 30, 2025

facebook-github-bot added fb-exported meta-exported labels Sep 30, 2025

cccclai force-pushed the export-D83606497 branch from 0918c6a to 74d7e98 Compare September 30, 2025 22:05

cccclai force-pushed the export-D83606497 branch from 74d7e98 to c6642fa Compare September 30, 2025 22:08

cccclai force-pushed the export-D83606497 branch from c6642fa to e1de9d0 Compare September 30, 2025 22:13

cccclai force-pushed the export-D83606497 branch from e1de9d0 to f2f84e4 Compare September 30, 2025 22:17

cccclai requested review from haowhsu-quic, shewu-quic and winskuo-quic September 30, 2025 22:18

haowhsu-quic reviewed Oct 2, 2025

View reviewed changes

backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py Show resolved Hide resolved

haowhsu-quic reviewed Oct 2, 2025

View reviewed changes

backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py Outdated Show resolved Hide resolved

haowhsu-quic reviewed Oct 2, 2025

View reviewed changes

cccclai force-pushed the export-D83606497 branch from f2f84e4 to ea4bebf Compare October 3, 2025 16:53

mergennachin reviewed Oct 3, 2025

View reviewed changes

backends/qualcomm/builders/op_argmax.py Outdated Show resolved Hide resolved

mergennachin reviewed Oct 3, 2025

View reviewed changes

backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py Outdated Show resolved Hide resolved

cccclai force-pushed the export-D83606497 branch from 3f825cf to af7ef91 Compare October 3, 2025 18:31

cccclai added ciflow/nightly and removed ciflow/nightly labels Oct 3, 2025

cccclai force-pushed the export-D83606497 branch 2 times, most recently from 214554c to 5077d00 Compare October 3, 2025 22:48

cccclai added ciflow/binaries ciflow/nightly and removed ciflow/binaries labels Oct 6, 2025

cccclai force-pushed the export-D83606497 branch from 5077d00 to a982ee6 Compare October 6, 2025 20:44

cccclai force-pushed the export-D83606497 branch from a982ee6 to 553be81 Compare October 7, 2025 16:58

haowhsu-quic reviewed Oct 7, 2025

View reviewed changes

backends/qualcomm/tests/test_qnn_delegate.py Outdated Show resolved Hide resolved

cccclai force-pushed the export-D83606497 branch from 553be81 to 4087c68 Compare October 7, 2025 17:07

haowhsu-quic approved these changes Oct 7, 2025

View reviewed changes

kimishpatel reviewed Oct 7, 2025

View reviewed changes

kimishpatel approved these changes Oct 7, 2025

View reviewed changes

cccclai changed the title ~~support argmax without dim kwargs~~ support argmax/argmin without dim kwargs and fix adaptive_max_pool3d Oct 7, 2025

cccclai force-pushed the export-D83606497 branch from 4087c68 to 3a5bff8 Compare October 7, 2025 17:50

cccclai merged commit e09abea into pytorch:main Oct 7, 2025
175 of 178 checks passed

pytorchbot mentioned this pull request Oct 7, 2025

[v1.0.0] Release Tracker #14288

Open

support argmax/argmin without dim kwargs and fix adaptive_max_pool3d #14710

support argmax/argmin without dim kwargs and fix adaptive_max_pool3d #14710

Uh oh!

Conversation

cccclai commented Sep 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Sep 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14710

❗ 1 Active SEVs

❌ 1 New Failure, 23 Pending

Uh oh!

facebook-github-bot commented Sep 30, 2025

Uh oh!

github-actions bot commented Sep 30, 2025

This PR needs a release notes: label

Uh oh!

facebook-github-bot commented Sep 30, 2025

Uh oh!

facebook-github-bot commented Sep 30, 2025

Uh oh!

facebook-github-bot commented Sep 30, 2025

Uh oh!

facebook-github-bot commented Sep 30, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

cccclai commented Oct 7, 2025

Uh oh!

meta-codesync bot commented Oct 7, 2025

Uh oh!

Uh oh!

haowhsu-quic left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

cccclai commented Oct 7, 2025

Uh oh!

pytorchbot commented Oct 7, 2025

Cherry picking #14710

Uh oh!

Uh oh!

cccclai commented Sep 30, 2025 •

edited

Loading

pytorch-bot bot commented Sep 30, 2025 •

edited

Loading

This PR needs a `release notes:` label