[PyTorch] Use FP16 tols for distributed tests with TF32 compute#1831
Merged
timmoon10 merged 6 commits intoNVIDIA:mainfrom Jun 19, 2025
Merged
[PyTorch] Use FP16 tols for distributed tests with TF32 compute#1831timmoon10 merged 6 commits intoNVIDIA:mainfrom
timmoon10 merged 6 commits intoNVIDIA:mainfrom
Conversation
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Collaborator
Author
|
/te-ci pytorch L1 |
Collaborator
Author
|
/te-ci pytorch L1 |
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Collaborator
Author
|
/te-ci pytorch L1 |
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch L1 |
KshitijLakhani
pushed a commit
that referenced
this pull request
Jun 27, 2025
* Use FP16 tols for tests with TF32 Signed-off-by: Tim Moon <tmoon@nvidia.com> * Use uniform init instead of constant init Signed-off-by: Tim Moon <tmoon@nvidia.com> * Revert constant init test, but reduce value Signed-off-by: Tim Moon <tmoon@nvidia.com> --------- Signed-off-by: Tim Moon <tmoon@nvidia.com> Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
chtruong814
pushed a commit
to chtruong814/TransformerEngine
that referenced
this pull request
Jul 17, 2025
…IA#1831) * Use FP16 tols for tests with TF32 Signed-off-by: Tim Moon <tmoon@nvidia.com> * Use uniform init instead of constant init Signed-off-by: Tim Moon <tmoon@nvidia.com> * Revert constant init test, but reduce value Signed-off-by: Tim Moon <tmoon@nvidia.com> --------- Signed-off-by: Tim Moon <tmoon@nvidia.com> Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
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
2 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.
Description
#1806 relaxed the numerical tolerances for the distributed tests using FP32 because they are actually doing compute in TF32. However, it seems the tolerances are still too tight because we are still seeing test failures on some systems. Since TF32 has the same number of mantissa bits as FP16, this PR drops the FP32 tolerances to match the FP16 tolerances in
torch.testing.assert_close.Type of change
Changes
Checklist: