Skip to content

Commit

Permalink
[XLA:GPU] Fix forward the flakiness of the test that was introduced i…
Browse files Browse the repository at this point in the history
…n the cl/678283878

The test is flaky because on the different platforms different versions win the competition in autotuner. I.e. from time to time cublas version wins because it is faster than triton version. At the same time both versions are fine.

PiperOrigin-RevId: 679005859
  • Loading branch information
loislo authored and Google-ML-Automation committed Sep 26, 2024
1 parent a61fd94 commit 3ca96df
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion xla/service/gpu/dot_algorithm_support_test.cc
Original file line number Diff line number Diff line change
Expand Up @@ -228,7 +228,7 @@ INSTANTIATE_TEST_SUITE_P(DotF32ForBf16Bf16F32Tests, DotAlgorithmSupportTest,
Combine(Values(PC::ALG_DOT_BF16_BF16_F32), Values(F32),
Values(F32), Values(CC(8, 0)),
Values(SemanticVersion{6, 0, 0}),
Values(BackendRestriction::kTritonOnly),
Values(BackendRestriction::kNoRestriction),
Values(Sizes{32, 32}, Sizes{16, 2})),
TestParamsToString);

Expand Down

0 comments on commit 3ca96df

Please sign in to comment.