[dynamo] Deprecate inline_inbuilt_nn_modules config by anijain2305 · Pull Request #178205 · pytorch/pytorch

anijain2305 · 2026-03-23T22:17:22Z

Stack from ghstack (oldest at bottom):

-> [dynamo] Deprecate inline_inbuilt_nn_modules config #178205

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben @jataylo @Lucaskabela @mlazos

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime [ghstack-poisoned]

pytorch-bot · 2026-03-23T22:17:25Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/178205

📄 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.

✅ You can merge normally! (4 Unrelated Failures)

As of commit 7211897 with merge base e534ad5 ():

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

inductor / inductor-test / test (inductor_torchbench, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (trunk failure)
vision_maskrcnn

UNSTABLE - The following jobs are marked as unstable, possibly due to flakiness on trunk:

inductor / inductor-cpu-test / test (cpu_inductor_torchbench, 1, 2, linux.2xlarge.amx, unstable) (gh) (#174929)
detectron2_maskrcnn_r_50_fpn
Lint OSDC (unstable) / Link checks / lint-urls / lint (gh)
Lint OSDC (unstable) / quick-checks / lint (gh)

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

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime ghstack-source-id: 2941144 Pull Request resolved: #178205

pytorch-bot · 2026-03-23T22:17:29Z

This PR needs a `release notes:` label

If your changes are user facing and intended to be a part of release notes, please use a label starting with release notes:.

If not, please add the topic: not user facing label.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "topic: not user facing"

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

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy kadeng muchulee8 amjames chauhang aakhundov coconutruben jataylo Lucaskabela [ghstack-poisoned]

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime ghstack-source-id: 8704e4f Pull Request resolved: #178205

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy kadeng muchulee8 amjames chauhang aakhundov coconutruben jataylo Lucaskabela [ghstack-poisoned]

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime ghstack-source-id: 5c85e7f Pull Request resolved: #178205

Lucaskabela · 2026-03-24T15:51:55Z

        if (
-            (
-                torch._dynamo.config.inline_inbuilt_nn_modules
-                or isinstance(self, variables.FSDPManagedNNModuleVariable)


Shouldn't we keep the check from the or?

Wait nvm since inline_inbuilt_nn_modules is true this entire portion is now true - checked myself sorry XD

No worries - I was confused too - Trust in Claude we must lol

Lucaskabela

Generally LGTM - minor comment on our deprecation message

Lucaskabela · 2026-03-24T15:52:35Z

    default=True,
    justknob="pytorch/compiler:inline_inbuilt_nn_modules",
+    deprecated=True,
+    deprecation_message="does not do anything, inline_inbuilt_nn_modules is always True",


Do we have an expected removal date? We could include that in the message if so

I dont want to add it now -- because in the past, this has been hard, so I am expecting reverts.

I can do this post 2.12 release, if the deprecate flag survives till 2.12 release.

Lucaskabela · 2026-03-24T15:55:02Z

        if (
-            (
-                torch._dynamo.config.inline_inbuilt_nn_modules
-                or isinstance(self, variables.FSDPManagedNNModuleVariable)


Wait nvm since inline_inbuilt_nn_modules is true this entire portion is now true - checked myself sorry XD

Skylion007 · 2026-03-24T16:08:33Z

@@ -910,13 +908,7 @@ def raise_local_type_error(obj: Any) -> NoReturn:


 def should_optimize_getattr_on_nn_module(value: Any) -> bool:


Return type should be typing_extensions.TypeIs[torch.nn.Module] to properly propogate typing info now.

anijain2305 · 2026-03-24T22:47:00Z

@pytorchbot merge

pytorchmergebot · 2026-03-24T22:49:23Z

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

Advanced Debugging

Check the merge workflow status
here

Summary: inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime X-link: pytorch/pytorch#178205 Approved by: https://github.com/Lucaskabela, https://github.com/mlazos Reviewed By: georgehong Differential Revision: D98065356 fbshipit-source-id: e55bae4fd2e0ff075f6a7de0215af05a18de5279

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime Pull Request resolved: #178205 Approved by: https://github.com/Lucaskabela, https://github.com/mlazos Co-authored-by: Xia-Weiwen <12522207+Xia-Weiwen@users.noreply.github.com>

inline_inbuilt_nn_modules has defaulted to True and the False path is legacy. This makes the config a no-op (always True), emits a deprecation warning when users set it to False, and removes all dead False-path code from runtime Pull Request resolved: pytorch#178205 Approved by: https://github.com/Lucaskabela, https://github.com/mlazos

…churn (#182531) PR #178205 stripped the cudagraph re-record limit entirely (made exceed_rerecord_limit a no-op). Restore it, but don't count StaticInputIdxMismatch (nn parameter / mark_static_address ptr churn under inline_inbuilt_nn_modules) toward the limit; structural reasons like CudagraphManagedIdxMismatch and ExpectedDeadIndicesBeforeGraphMismatch do count. Authored with assistance from Claude. Differential Revision: [D103898330](https://our.internmc.facebook.com/intern/diff/D103898330) Pull Request resolved: #182531 Approved by: https://github.com/BoyuanFeng ghstack dependencies: #182524

…churn (pytorch#182531) PR pytorch#178205 stripped the cudagraph re-record limit entirely (made exceed_rerecord_limit a no-op). Restore it, but don't count StaticInputIdxMismatch (nn parameter / mark_static_address ptr churn under inline_inbuilt_nn_modules) toward the limit; structural reasons like CudagraphManagedIdxMismatch and ExpectedDeadIndicesBeforeGraphMismatch do count. Authored with assistance from Claude. Differential Revision: [D103898330](https://our.internmc.facebook.com/intern/diff/D103898330) Pull Request resolved: pytorch#182531 Approved by: https://github.com/BoyuanFeng ghstack dependencies: pytorch#182524

anijain2305 requested review from aorenste and zou3519 as code owners March 23, 2026 22:17

pytorch-bot Bot added ciflow/inductor ciflow/torchtitan Run TorchTitan integration tests module: dynamo module: inductor labels Mar 23, 2026

anijain2305 mentioned this pull request Mar 23, 2026

[dynamo] Adjust more tests to prepare for inline_inbuilt_nn_module deprecation #178067

Closed

anijain2305 added ciflow/trunk Trigger trunk jobs on your pull request release notes: dynamo labels Mar 24, 2026

anijain2305 requested review from Lucaskabela, guilhermeleobas, mlazos and williamwen42 March 24, 2026 05:10

Lucaskabela reviewed Mar 24, 2026

View reviewed changes

Lucaskabela approved these changes Mar 24, 2026

View reviewed changes

Skylion007 reviewed Mar 24, 2026

View reviewed changes

mlazos approved these changes Mar 24, 2026

View reviewed changes

pytorchmergebot added the merging label Mar 24, 2026

pytorchmergebot closed this in d10d6a1 Mar 24, 2026

pytorchmergebot added Merged and removed merging labels Mar 24, 2026

github-actions Bot deleted the gh/anijain2305/1109/head branch April 24, 2026 02:25

eellison mentioned this pull request May 5, 2026

[inductor] gate cudagraph re-record limit: don't count parameter ptr churn #182531

Closed

		@@ -910,13 +908,7 @@ def raise_local_type_error(obj: Any) -> NoReturn:


		def should_optimize_getattr_on_nn_module(value: Any) -> bool:

Conversation

anijain2305 commented Mar 23, 2026 • edited by pytorch-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Mar 23, 2026 • 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/178205

✅ You can merge normally! (4 Unrelated Failures)

Uh oh!

pytorch-bot Bot commented Mar 23, 2026

This PR needs a release notes: label

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!

Lucaskabela 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!

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!

anijain2305 commented Mar 24, 2026

Uh oh!

pytorchmergebot commented Mar 24, 2026

Merge started

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

anijain2305 commented Mar 23, 2026 •

edited by pytorch-bot Bot

Loading

pytorch-bot Bot commented Mar 23, 2026 •

edited

Loading

This PR needs a `release notes:` label