[Performance] Error encountered with the SkipLayerNormalization node.

[DakeQQ]

Describe the issue

If a GPT-like ONNX model contains SkipLayerNormalization nodes, an error occurs during loading with the OpenVINOExecutionProvider; however, the model works well with other ExecutionProviders, such as CPU, CUDA, and DML. The following is the error message:

line 561, in _create_inference_session
    sess.initialize_session(providers, provider_options, disabled_optimizers)
    ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Exception during initialization: /onnxruntime/onnxruntime/core/providers/openvino/ov_interface.cc:79 std::shared_ptr<ov::Model> onnxruntime::openvino_ep::OVCore::ReadModel(std::string&&, const std::string&) [OpenVINO-EP] [OpenVINO-EP] Exception while Reading network: Check 'onnx_node.get_outputs_size() <= outputs_size' failed at src/frontends/onnx/frontend/src/core/graph.cpp:392:
FrontEnd API failed with GeneralFailure:
Expected output number of SkipLayerNormalization node is 4 while the implementation provides 1 outputs

To reproduce

import onnxruntime

onnx_model_A = "Whisper_Encoder.onnx" 
provider_options = [
        {
            'device_type': 'CPU',                        
            'precision': 'ACCURACY',                      
            'num_of_threads': 1,
            'num_streams': 1,
            'enable_opencl_throttling': True,
            'enable_qdq_optimizer': False,                
            'disable_dynamic_shapes': False
        }
    ]

session_opts = onnxruntime.SessionOptions()
session_opts.log_severity_level = 0

ort_session_A = onnxruntime.InferenceSession(onnx_model_A, sess_options=session_opts, providers=['OpenVINOExecutionProvider'], provider_options=provider_options)

Urgency

No response

Platform

Linux

OS Version

Ubuntu 24.04

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.22.1

ONNX Runtime API

Python

Architecture

X64

Execution Provider

OpenVINO

Execution Provider Library Version

1.22.0

Model File

Whisper_Encoder.zip

Is this a quantized model?

No