Are exported onnx files encrypted?

Tao 5.x has hdf5 pretrained models, as told in this forum post: https://forums.developer.nvidia.com/t/what-is-hdf5-files/274322.

When we train using them, we get hdf5 files. And when we export, we get onnx file.

Now, can you please tell if this onnx file is encrypted? (for example, etlt file is encrypted with the key that we give in the exoort command)

The onnx file is not encrypted.

Thanks for the answer. As it is also mentioned in the same forum post, that, etlt is being deprecated, can we say that encrypting the models is being removed, or, is there a way to encrypt the onnx models, or convert them to etlt?

Yes.

You can refer to tao_deploy/internal/encrypt_onnx.py at 31c7e0ed3fe48942c254b3b85517e7418eea17b3 · NVIDIA/tao_deploy · GitHub.

Thank you very much.

Please tell if there is a way to run this script via launcher. Or any other way.

I have been trying to create a venv with the requirements.txt in the docker sub folder, but running into dependency issues (mainly with eff, eff_tao_encryption, scipy). Tried with python 3.8, 3.9 and 3.10 and mkvirtualenv.

You can run with tao-deploy docker.

Would you mind giving an example?

This documentation states that when we run the command starting with tao deploy, the required container is downloaded automatically and instantiated.

I tried figuring out using command line args help, but couldn’t find a way.

Please use docker run to trigger docker.
nvcr.io/nvidia/tao/tao-toolkit:5.3.0-deploy
TAO Toolkit | NVIDIA NGC.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.