pytorch ortmodule examples not running on Ubuntu 20.04?

[shibdas]

Hi,
I'm in the process of writing an execution provider for ONNX. While the ORT inference works well with the new EP, I'm finding it very difficult to run the training examples like https://github.com/pytorch/ort/blob/main/tests/bert_for_sequence_classification.py on ubuntu 20.04. I'm able to run that example fine using the docker but I'm getting different errors while running it locally using the pytorch and cuda version mentioned in the repository. I've tried several python environments to match the docker environment and still hit this error while running python bert_for_sequence_classification.py.
"Inconsistency detected by ld.so: dl-version.c: 205: _dl_check_map_versions: Assertion `needed != NULL' failed!"
Running with LD_DEBUG=all, I saw that it's looking at some gcc and glibc versions which are not there in ubuntu 20.04 but present in the ubuntu 18.04 docker image. I stopped as replacing the libgcc would stop other programs to work.

Looking at the CI build runs, it seems they are also running it on docker with Ubuntu 18.04 or centos as far as I can tell. My question is, has anyone been able to run the example on ubuntu 20,04. I need to ideally run them with the local build of onnxruntime but first I'd like to just run the examples with released version of ORT and torch_ort.

Thanks for your help!