[export] backed_size_oblivious tutorial #3400
Open
+25
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/3400
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 5b1689e with merge base ab2aafd ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pianpwk
requested review from
avikchaudhuri,
angelayi,
tugsbayasgalan,
bobrenjc93 and
laithsakka
angelayi
reviewed
Jun 26, 2025
| @@ -591,6 +592,30 @@ def forward(self, x, y): | |||
| "bool_val": None, | |||
| } | |||
|
|
|||
| ###################################################################### | |||
| # (experimental) Avoiding 0/1 specialization | |||
There was a problem hiding this comment.
should this be added after the section on 0/1 specialization? or move the section on 0/1 specialization to right before here?
| # (experimental) Avoiding 0/1 specialization | ||
| # ^^^^^^^^^^^^^^^^^^ | ||
| # | ||
| # Export provides an experimental option to avoid specializing on size 0/1 sample inputs. Users can turn on `torch.fx.experimental._config.backed_size_oblivious = True` to enable this behavior. |
There was a problem hiding this comment.
Suggested change
| # Export provides an experimental option to avoid specializing on size 0/1 sample inputs. Users can turn on `torch.fx.experimental._config.backed_size_oblivious = True` to enable this behavior. | |
| # Export provides an experimental option to avoid specializing on size 0/1 sample inputs. Users can turn on ``torch.fx.experimental._config.backed_size_oblivious = True`` to enable this behavior. |
| y = torch.randn(1) | ||
| dynamic_shapes = {"x": (Dim.AUTO, Dim.AUTO), "y": (Dim.AUTO,)} | ||
| with torch.fx.experimental._config.patch(backed_size_oblivious=True): | ||
| ep = export(Foo(), (x, y), dynamic_shapes=dynamic_shapes) |
| ep = export(Foo(), (x, y), dynamic_shapes=dynamic_shapes) | ||
| ep.module()(torch.randn(8, 1), torch.randn(1)) | ||
| ep.module()(torch.randn(5, 6), torch.randn(6)) | ||
|
|
There was a problem hiding this comment.
i wonder if it would be good to add an example where the behavior is weir after setting backed_size_oblivious. your example here is a little simple.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #ISSUE_NUMBER
Description
Adds an export tutorial section for avoiding 0/1 specialization
Checklist
cc @avikchaudhuri @gmagogsfm @zhxchen17 @tugsbayasgalan @angelayi @suo @ydwu4