I think it makes sense to stick to one opset version (you can specify it in PyTorch when exporting). For this, I suggest we mention this in the README and generate a warning in Deeploy if it does not match. In any case, I do not think it makes sense to duplicate here.

@Victor-Jung, @lukamac, @federicobrancasi, do you have a suggestion/preference on which version to use?

Originally posted by @Xeratec in #92 (comment)