AIMIGRAPHX-824 Set basic graph optimization for ONNX sessions in accuracy_checker.py

[kentqian]

Motivation

These files cannot be executed correctly during CI

• bert_base_cased_1_fp16_gpu.onnx
• bert_large_uncased_1_fp16_gpu.onnx
• distilgpt2_1_fp16_gpu.onnx

Here is Error output:

Traceback (most recent call last):
File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 359, in
main()
File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 278, in main
sess = ort.InferenceSession(model_name,
File "/usr/local/lib/python3.10/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 485, in init
self._create_inference_session(providers, provider_options, disabled_optimizers)
File "/usr/local/lib/python3.10/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 584, 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_src/onnxruntime/core/graph/graph_utils.cc:29 int onnxruntime::graph_utils::GetIndexFromName(const onnxruntime::Node&, const std::string&, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_onnx::Pow_1551for node: Mul_33/SimplifiedLayerNormFusion/

Technical Details

This defect is related to the level of graph optimization when the level setting is as default or ORT_ENABLE_ALL, since SimplifiedLayerNormFusion will be triggered by ORT_ENABLE_ALL setting. This is a workaround as replacing ORT_ENABLE_ALL
to ORT_ENABLE_BASIC. The latest ORT(1.23.2) via pip is impacted.
One issue thread on Microsoft’s OnnxRT public repro had been created for this similar issue. #26717 Microsoft's OnnxRT

[Copilot]

Pull request overview

This PR updates the ONNXRuntime session configuration used by tools/accuracy/accuracy_checker.py to avoid a known ONNXRuntime crash triggered by higher graph optimization levels (e.g., ORT_ENABLE_ALL) on certain transformer ONNX models during CI.

Changes:

• Set onnxruntime.SessionOptions.graph_optimization_level to ORT_ENABLE_BASIC for ONNX sessions created by the accuracy checker.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.