Arm backend: relax tolerance for test_constant_pad_nd_vgf_quant#19263
Arm backend: relax tolerance for test_constant_pad_nd_vgf_quant#19263rascani wants to merge 1 commit intopytorch:mainfrom
Conversation
The VGF + INT8 quantized pipeline for constant_pad_nd hits the same randomized-input numerical drift the TOSA INT variant has been working around since MLETORCH-989: the test_data_suite uses unseeded torch.rand so unfortunate input draws produce a single-element max abs diff (~6e-3) that exceeds the default atol (~4e-3). Apply the same atol/rtol the TOSA INT variant uses (atol=0.005, rtol=0.01) so the VGF run shares the existing umbrella ticket rather than continuing to flake on the same underlying issue. Authored with Claude Code.
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19263
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 Awaiting Approval, 2 New Failures, 4 Unrelated FailuresAs of commit 333a8db with merge base e84a418 ( NEW FAILURES - The following jobs have failed:
FLAKY - The following job failed but was likely due to flakiness present on trunk:
BROKEN TRUNK - The following jobs failed but was present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This PR needs a
|
Summary
The VGF + INT8 quantized pipeline for constant_pad_nd hits the same randomized-input numerical drift the TOSA INT variant has been working around since MLETORCH-989: the test_data_suite uses unseeded torch.rand so unfortunate input draws produce a single-element max abs diff (~6e-3) that exceeds the default atol (~4e-3). Apply the same atol/rtol the TOSA INT variant uses (atol=0.005, rtol=0.01) so the VGF run shares the existing umbrella ticket rather than continuing to flake on the same underlying issue.
Authored with Claude Code.
Test plan
CI
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell