Warning: Checker does not support models with experimental ops: ATen about ort HOT 2 CLOSED

you can install onnxruntime-training for cuda 11.5 using the command:

python -m pip install onnxruntime-training -f https://download.onnxruntime.ai/onnxruntime_stable_cu115.html

With onnxruntime-training for cuda 11.5, I think you should not see that error.

from ort.

Can you share the cuda version of onnxruntime-training and torch installed? From the information you provided, the torch cuda version is 11.6 and my guess is that the onnxruntime-training cuda version is 10.2 (which is the reason for the error I think).

The cuda version for the two should align.

python -c "from onnxruntime.capi import build_and_package_info as ort_info; print(ort_info.cuda_version)"
python -c "import torch; print(torch.version.cuda)"

from ort.