Allow NJT by default for weights_only torch.load #140304

mikaylagawarecki · 2024-11-11T20:58:11Z

Stack from ghstack (oldest at bottom):

-> Allow NJT by default for weights_only torch.load #140304

[ghstack-poisoned]

pytorch-bot · 2024-11-11T20:58:14Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/140304

📄 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 or our office hours

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:

GLIBC not found in Nova workflows

✅ No Failures

As of commit 70c9d2e with merge base 057f0dc ():
💚 Looks good so far! There are no failures yet. 💚

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

ghstack-source-id: aae64ce Pull Request resolved: #140304

jbschlosser

LGTM thanks!

mikaylagawarecki · 2024-11-11T23:45:02Z

@pytorchbot merge

pytorchmergebot · 2024-11-11T23:47:06Z

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

huydhn · 2024-11-12T04:51:31Z

@pytorchbot revert -m 'A bunch of failure shows up in trunk after this lands, so probably a landrace' -c landrace

https://hud.pytorch.org/pytorch/pytorch/commit/455dc4c14264a0cd7d70ba5328382a9fb7769094

pytorchmergebot · 2024-11-12T04:52:59Z

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

This reverts commit 455dc4c. Reverted #140304 on behalf of https://github.com/huydhn due to A bunch of failure shows up in trunk after this lands, so probably a landrace ([comment](#140304 (comment)))

pytorchmergebot · 2024-11-12T04:53:13Z

@mikaylagawarecki your PR has been successfully reverted.

[ghstack-poisoned]

ghstack-source-id: fd9aecc Pull Request resolved: #140304

mikaylagawarecki · 2024-11-12T20:14:36Z

@pytorchbot merge

pytorchmergebot · 2024-11-12T20:19:17Z

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

#140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?)** Further, I am wrapping all these imports in functions that will only be called if the GLOBAL is found in the checkpoint to avoid all users of `torch.load(weights_only=True)` from always paying the cost of importing these Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

…h.load (take 2)" Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

…h.load (take 2)" Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

…h.load (take 2)" Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

…h.load (take 2)" Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

Per discussion with malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]

@malfet

Per discussion with @malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of #140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) @jbschlosser Pull Request resolved: #140739 Approved by: https://github.com/malfet

mikaylagawarecki · 2024-11-21T19:36:55Z

Relanded in #140739

Pull Request resolved: pytorch#140304 Approved by: https://github.com/jbschlosser

)" This reverts commit 455dc4c. Reverted pytorch#140304 on behalf of https://github.com/huydhn due to A bunch of failure shows up in trunk after this lands, so probably a landrace ([comment](pytorch#140304 (comment)))

Pull Request resolved: pytorch#140304 Approved by: https://github.com/jbschlosser

)" This reverts commit 1f28235. Reverted pytorch#140304 on behalf of https://github.com/mikaylagawarecki due to Breaking internal tests due to missing torch.nested._internal ([comment](pytorch#140304 (comment)))

@malfet

…0739) Per discussion with @malfet, only allow weights_only unpickler to load NJT if `torch.nested` and `torch._dynamo` are imported (this is slightly weird as technically `torch.nested` is actually imported by default and `torch._dynamo.decorators._DimRange` is actually what needs to be imported) we can't import this from `torch.nested` as this would - undo dynamo lazy import - cause circular import =========================== Redo of pytorch#140304 caused issues as `torch.nested._internal.foo` needs to be imported, which causes issues like ```python torch/_weights_only_unpickler.py", line 339, in load if full_path in _get_allowed_globals(): torch/_weights_only_unpickler.py", line 188, in _get_allowed_globals torch.nested._internal.nested_tensor.NestedTensor AttributeError: module 'torch.nested' has no attribute '_internal' ``` **This likely wasn't caught in our CI because imports are global during unit tests(?), so we use subprocess to properly test this time** Differential Revision: [D65961691](https://our.internmc.facebook.com/intern/diff/D65961691) @jbschlosser Pull Request resolved: pytorch#140739 Approved by: https://github.com/malfet

Allow nested tensor by default for weights_only torch.load

42cbdd2

[ghstack-poisoned]

mikaylagawarecki added a commit that referenced this pull request Nov 11, 2024

Allow nested tensor by default for weights_only torch.load

337a8ca

ghstack-source-id: aae64ce Pull Request resolved: #140304

mikaylagawarecki requested a review from jbschlosser November 11, 2024 21:03

mikaylagawarecki added release notes: python_frontend python frontend release notes category topic: improvements topic category labels Nov 11, 2024

mikaylagawarecki changed the title ~~Allow nested tensor by default for weights_only torch.load~~ Allow NJT by default for weights_only torch.load Nov 11, 2024

jbschlosser approved these changes Nov 11, 2024

View reviewed changes

mikaylagawarecki marked this pull request as ready for review November 11, 2024 22:07

pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label Nov 11, 2024

pytorchmergebot added the merging label Nov 11, 2024

pytorchmergebot added the Merged label Nov 12, 2024

pytorchmergebot closed this in 455dc4c Nov 12, 2024

pytorchmergebot removed the merging label Nov 12, 2024

pytorchmergebot added Reverted ci-no-td Do not run TD on this PR labels Nov 12, 2024

pytorchmergebot reopened this Nov 12, 2024

Update on "Allow NJT by default for weights_only torch.load"

70c9d2e

[ghstack-poisoned]

mikaylagawarecki added a commit that referenced this pull request Nov 12, 2024

Allow nested tensor by default for weights_only torch.load

633762b

ghstack-source-id: fd9aecc Pull Request resolved: #140304

pytorchmergebot added the merging label Nov 12, 2024

mikaylagawarecki closed this Nov 21, 2024

github-actions bot deleted the gh/mikaylagawarecki/286/head branch December 22, 2024 02:09

Archana-Shinde1 mentioned this pull request May 5, 2025

[REQUEST] Backport fix for Critical CVE-2025-32434 (9.8) to PyTorch v2.1.2 open-ce/open-ce#1362

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Allow NJT by default for weights_only torch.load #140304

Allow NJT by default for weights_only torch.load #140304

Uh oh!

mikaylagawarecki commented Nov 11, 2024 •

edited

Loading

Uh oh!

pytorch-bot bot commented Nov 11, 2024 •

edited

Loading

Uh oh!

jbschlosser left a comment

Uh oh!

mikaylagawarecki commented Nov 11, 2024

Uh oh!

pytorchmergebot commented Nov 11, 2024

Uh oh!

huydhn commented Nov 12, 2024

Uh oh!

pytorchmergebot commented Nov 12, 2024

Uh oh!

pytorchmergebot commented Nov 12, 2024

Uh oh!

mikaylagawarecki commented Nov 12, 2024

Uh oh!

pytorchmergebot commented Nov 12, 2024

Uh oh!

mikaylagawarecki commented Nov 21, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Allow NJT by default for weights_only torch.load #140304

Allow NJT by default for weights_only torch.load #140304

Uh oh!

Conversation

mikaylagawarecki commented Nov 11, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Nov 11, 2024 • 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/140304

❗ 1 Active SEVs

✅ No Failures

Uh oh!

jbschlosser left a comment

Choose a reason for hiding this comment

Uh oh!

mikaylagawarecki commented Nov 11, 2024

Uh oh!

pytorchmergebot commented Nov 11, 2024

Merge started

Uh oh!

huydhn commented Nov 12, 2024

Uh oh!

pytorchmergebot commented Nov 12, 2024

Uh oh!

pytorchmergebot commented Nov 12, 2024

Uh oh!

mikaylagawarecki commented Nov 12, 2024

Uh oh!

pytorchmergebot commented Nov 12, 2024

Merge started

Uh oh!

mikaylagawarecki commented Nov 21, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

mikaylagawarecki commented Nov 11, 2024 •

edited

Loading

pytorch-bot bot commented Nov 11, 2024 •

edited

Loading