Certain ONNX Operators

[ar-roy]

Hi team, I've noticed that ONNX operators have been slowly added over previous releases this past year (kudos!).

There are two operators missing when attempting to convert a pretrained torch ResNet18 model via compile_torch_model:

1. Expand
2. ReduceL2

Are these two operators missing due to limitations on the FHE side or because they are not commonly used?
Thanks!

[fd0r]

Hello @ar-roy, thanks for your interest in Concrete ML!

These two operators are not supported at the moment because we didn't find a use for them yet but we will add them to our roadmap!

If you can't wait for us to implement them you can also add them yourselves and do a PR to this repository. If you want to do so I would advise you to start looking at our QuantizedOps objects in https://github.com/zama-ai/concrete-ml/blob/main/src/concrete/ml/quantization/quantized_ops.py

We'll ping you here once the feature is available in the latest release. 🙂

[ar-roy]

Thanks @fd0r!

I'll look into doing a PR if necessary or see if there is some work-around for the operations on my end.

I'll close this out for now, looking forward to new releases as they come! 👍