add and fix OpInfo tests for the default partitioner#165372
Closed
bdhirsh wants to merge 6 commits intogh/bdhirsh/674/basefrom
Closed
add and fix OpInfo tests for the default partitioner#165372bdhirsh wants to merge 6 commits intogh/bdhirsh/674/basefrom
bdhirsh wants to merge 6 commits intogh/bdhirsh/674/basefrom
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/165372
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 90343a5 with merge base e787d53 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests [ghstack-poisoned]
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests [ghstack-poisoned]
ezyang
reviewed
Oct 14, 2025
| elif ( | ||
| "tensor_meta" not in node.meta | ||
| and node.op == "call_function" | ||
| and not isinstance(node.meta.get("val"), torch._subclasses.FakeTensor) |
ezyang
approved these changes
Oct 14, 2025
Collaborator
Author
|
@pytorchbot merge |
Collaborator
Merge startedYour 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 |
zhudada0120
pushed a commit
to zhudada0120/pytorch
that referenced
this pull request
Oct 15, 2025
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests Pull Request resolved: pytorch#165372 Approved by: https://github.com/ezyang ghstack dependencies: pytorch#165327
Contributor
|
@pytorchbot revert -m "Looks like it broke slow jobs, see https://hud.pytorch.org/hud/pytorch/pytorch/331b7cc054415210ec73f4e7e4571f8a0c21ed62/1?per_page=50&name_filter=slow&mergeEphemeralLF=true" -c nosignal |
Collaborator
|
@pytorchbot successfully started a revert job. Check the current status here. |
Collaborator
|
@bdhirsh your PR has been successfully reverted. |
pytorchmergebot
added a commit
that referenced
this pull request
Oct 15, 2025
This reverts commit bcfea48. Reverted #165372 on behalf of https://github.com/malfet due to Looks like it broke slow jobs, see https://hud.pytorch.org/hud/pytorch/pytorch/331b7cc054415210ec73f4e7e4571f8a0c21ed62/1?per_page=50&name_filter=slow&mergeEphemeralLF=true ([comment](#165372 (comment)))
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests [ghstack-poisoned]
bdhirsh
commented
Oct 15, 2025
test/functorch/test_aotdispatch.py
Outdated
| decorator=toleranceOverride({torch.float32: tol(atol=1e-05, rtol=1e-05)}), | ||
| # This delta is coming entirely from the clone() on tangents | ||
| # in AOTDispatcher to make them contiguous | ||
| decorator=toleranceOverride({torch.float32: tol(atol=4e-05, rtol=1e-05)}), |
Collaborator
Author
There was a problem hiding this comment.
fyi the slow-test failure that caused the revert was interesting:
(1) this test wobbled tolerance a bit:
PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=32 PYTORCH_TEST_WITH_SLOW=1 PYTORCH_TEST_SKIP_FAST=1 python test/functorch/test_aotdispatch.py TestEagerFusionOpInfoCPU.test_aot_autograd_symbolic_default_partition_exhaustive_linalg_pinv_singular_cpu_float32
(2) I used @SherlockNoMad 's handy DebugMode to find the different kernels running between eager and aot_eager, and I found the difference is coming entirely from AOTDispatcher emitting clone() on tangents in the backward (presumably this causes us to call linalg.pinv backward with different striding, and the op's numerics are sensitive to strides)
We're going to have to deal with this in the "bitwise equality" workstream. I dont' want to deal with it in this PR, but we might want to consider not running the clone at all, and raising an error if we got the strides of our tangents wrong / require the user to tell us what striding they want for the tangents.
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests [ghstack-poisoned]
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests [ghstack-poisoned]
Collaborator
|
Starting merge as part of PR stack under #164577 |
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 16, 2025
I'm cleaning this PR up as a proper way of disabling functionalization via config in AOTDispatcher. I removed the non-functionalization related changes from the original version: (1) preventing proxy mode (and functionalization) from incorrectly decomposing CIA ops (Ed has a PR for it here: #164939) (2) preventing python-dispatcher-based decomps above autograd from running. I'm not doing this for now, will likely do it in a followup Pull Request resolved: #164577 Approved by: https://github.com/ezyang ghstack dependencies: #165372
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Oct 21, 2025
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests Pull Request resolved: pytorch#165372 Approved by: https://github.com/ezyang ghstack dependencies: pytorch#165327
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Oct 21, 2025
…#165372)" This reverts commit bcfea48. Reverted pytorch#165372 on behalf of https://github.com/malfet due to Looks like it broke slow jobs, see https://hud.pytorch.org/hud/pytorch/pytorch/331b7cc054415210ec73f4e7e4571f8a0c21ed62/1?per_page=50&name_filter=slow&mergeEphemeralLF=true ([comment](pytorch#165372 (comment)))
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Oct 21, 2025
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests Pull Request resolved: pytorch#165372 Approved by: https://github.com/ezyang
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Oct 21, 2025
…#164577) I'm cleaning this PR up as a proper way of disabling functionalization via config in AOTDispatcher. I removed the non-functionalization related changes from the original version: (1) preventing proxy mode (and functionalization) from incorrectly decomposing CIA ops (Ed has a PR for it here: pytorch#164939) (2) preventing python-dispatcher-based decomps above autograd from running. I'm not doing this for now, will likely do it in a followup Pull Request resolved: pytorch#164577 Approved by: https://github.com/ezyang ghstack dependencies: pytorch#165372
zhudada0120
pushed a commit
to zhudada0120/pytorch
that referenced
this pull request
Oct 22, 2025
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests Pull Request resolved: pytorch#165372 Approved by: https://github.com/ezyang
zhudada0120
pushed a commit
to zhudada0120/pytorch
that referenced
this pull request
Oct 22, 2025
…#164577) I'm cleaning this PR up as a proper way of disabling functionalization via config in AOTDispatcher. I removed the non-functionalization related changes from the original version: (1) preventing proxy mode (and functionalization) from incorrectly decomposing CIA ops (Ed has a PR for it here: pytorch#164939) (2) preventing python-dispatcher-based decomps above autograd from running. I'm not doing this for now, will likely do it in a followup Pull Request resolved: pytorch#164577 Approved by: https://github.com/ezyang ghstack dependencies: pytorch#165372
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests
Stack from ghstack (oldest at bottom):