test: loosening absolute tolerance for layer norm tests

[laserkelvin]

PhysicsNeMo Pull Request

Description

This PR modifies the absolute numerical tolerance used to compare fused and unfused implementations of the equivariant layer norm, which triggers on certain platforms.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• The CHANGELOG.md is up to date with these changes.
• An issue is linked to this pull request.
• If I am implementing a new model or modifying any existing model, I have followed the Models Implementation Coding Standards.

Dependencies

N/A

Review Process

All PRs are reviewed by the PhysicsNeMo team before merging.

Depending on which files are changed, GitHub may automatically assign a maintainer for review.

We are also testing AI-based code review tools (e.g., Greptile), which may add automated comments with a confidence score.
This score reflects the AI’s assessment of merge readiness and is not a qualitative judgment of your work, nor is
it an indication that the PR will be accepted / rejected.

AI-generated feedback should be reviewed critically for usefulness.
You are not required to respond to every AI comment, but they are intended to help both authors and reviewers.
Please react to Greptile comments with 👍 or 👎 to provide feedback on their accuracy.

[greptile-apps]

Greptile Summary

This PR increases the absolute tolerance for torch.float32 equivariance tests from 1e-4 to 5e-3 (a 50× relaxation) in the shared get_rtol_atol helper to accommodate platform-specific numerical differences in the fused layer norm. The change is minimal and the motivation is clear, but the new atol now equals the float16 threshold, and the Notes docstring still claims float32 delivers "~4-5 decimal digits" of precision.

Important Files Changed

Filename	Overview
test/experimental/nn/symmetry/conftest.py	Loosens float32 atol from 1e-4 to 5e-3 (50×) for layer norm equivariance tests; docstring precision claim now inconsistent with the new value.

Comments Outside Diff (1)

1. test/experimental/nn/symmetry/conftest.py, line 95-96 (link)

   Docstring precision claim inconsistent with new atol

   The Notes section still documents float32 as providing ~4-5 decimal digits of precision, but atol=5e-3 corresponds to roughly 2-3 decimal digits — the same level as float16. Consider updating the comment to reflect the actual tolerance in use.

_{Reviews (1): Last reviewed commit: "test: loosening absolute tolerance for l..." | Re-trigger Greptile}

[laserkelvin]

/blossom-ci

[laserkelvin]

GLOBE related tests fail on GitHub CI - paging Dr. @peterdsharpe!

[peterdsharpe]

@laserkelvin Aware of this, seems to be sporadically failing - apologies! Investigating here: #1580

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[laserkelvin]

/blossom-ci