Ti-Tai Wang

**1. Is strict_mode necessary for onnxruntime?** onnxruntime doesn't run strict mode check, but it simply extends shape type inference results from ONNX. That's why this invalid model can't be executed...

> @AllenTiTaiWang a few questions that are not explained on the PR description > > a) What are the "serious" type inference error? list all, preferably with examples b) Is...

Thanks for taking a look. I think I might mess up the CUDA version between these two. I will check again this week.

I keep the onnx, and use it to run with onnxruntime==1.12.1 and nightly onnxruntime (built with CUDA 11.6) [layer_norm.zip](https://github.com/microsoft/onnxruntime/files/9501174/layer_norm.zip) The result doesn't algin. This is how I built it: ```bash...

cc @justinchuby if you have more insight to provide.

Indeed, without --enable_training, the issue is solved. Is the mismatch with --enable_training is an expected behavior? Why is there a needed difference?

> I didn't understand the goal of this PR. Why do we need this set of scripts? Is this for creating custom ops or just an example for ORT? Isnt...

> @AllenTiTaiWang, thanks for the effort. > > Could you add license header to py files? Done > Did you able to test the accuracy of exported model (like compare...

> > > @AllenTiTaiWang, thanks for the effort. > > > Could you add license header to py files? > > > > > > Done > > > Did...

> I saw different results from PyTorch and ORT. For example, the stdout from `python export.py -m facebook/bart-base`. I used PyTorch 1.12.1+cu116, transformers 4.18.0 and onnxruntime-gpu 1.12.1: > > ```...