LayerNormalization doesnt' work as expected in MAC.

[fatcat-z]

Describe the issue

After converted a PyTorch model to ONNX model, the inferences results of these 2 models are same in CUDA/CPU environment. But in MAC, the result of converted ONNX model is different then PyTorch model's result.

To reproduce

native_layer_norm_issue_size_1.zip

Execute repro_issue.py file in attached zip file in CPU and MAC environments, you can see the results.

The expected results: The file should PASS in both environments.
Actual results: It fails in MAC environment.

Urgency

No response

Platform

Mac

OS Version

mac OS

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.17.3

ONNX Runtime API

Python

Architecture

Other / Unknown

Execution Provider

Default CPU

Execution Provider Library Version

No response