Hello, I'm Jongmin Kim, a student studying DNN accelerator, and I have been interested in your simulator.

According to the README, the following command should generate an ONNX graph for GPT-2:
python3 ./scripts/generate_transformer_onnx.py --model gpt2

However, it results in the following error message:

File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 2007, in _run_symbolic_function
raise errors.UnsupportedOperatorError(
torch.onnx.errors.UnsupportedOperatorError: Exporting the operator 'aten::scaled_dot_product_attention' to ONNX opset version 11 is not supported. Support for this operator was added in version 14, try exporting with this version.
Traceback (most recent call last):

I have verified that the version is 22, and I don’t believe this is the issue, as the latest versions of onnx and onnxruntime are installed, as specified in the Dockerfile.

I am currently attempting to export ONNX graphs for GPT-2, Llama 3, and OPT using the ONNX optimizer (AutoModelForXXX, ORTModelForXXX, etc.).
However, I’m wondering if there are any simpler or more efficient methods to resolve this issue.

Thank you for your time, and I look forward to your guidance.