[dynamo] Add guards for deterministic algos #96695
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/96695
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 4d784f5: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
colesbury
changed the title
deterministic algos.
colesbury
force-pushed
the
issue93537
branch
2 times, most recently
from
ef07a06
to
71c2849
Compare
Inductor now falls back to eager mode for deterministic algos. Add guards in dynamo to check if the deterministic algos mode changes. See pytorch#93537
|
@ngimel can you review this when you get a chance? |
test/inductor/test_torchinductor.py
Outdated
| r1 = fn(idx, values) | ||
| for _ in range(10): | ||
| rn = fn(idx, values) | ||
| assert (r1 == rn).all() |
There was a problem hiding this comment.
should it be self.assertEqual with 0 tolerance?
There was a problem hiding this comment.
I'll update it.
Is there a functional difference? (I originally used assert because that's what was used by the nearby test cases.)
There was a problem hiding this comment.
No functional difference, TestCase.assertEqual provides more detailed error message when it doesn't pass (maximum difference, index of the maximum difference etc).
|
@pytorchbot merge |
Merge failedReason: This PR needs a label If not, please add the To add a label, you can comment to pytorchbot, for example For more information, see Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot merge |
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 |
Inductor now falls back to eager mode for deterministic algos. Add guards in dynamo to check if the deterministic algos mode changes.
See #93537
cc @soumith @voznesenskym @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @desertfire