Justin Chu

Will this actually drop the `...` and the `k` dimensions?

According to https://github.com/onnx/onnx/pull/6010#discussion_r1535079663, Every occurrence of ellipsis must represent the same number of dimensions. Here The first `...` is two but the second `...` is zero.

Thank you for the improvement! Could you help adding context for reviewers by improving the pull request title and description?

Could you follow https://github.com/onnx/onnx/blob/main/CONTRIBUTING.md#coding-style to fix lint errors reported in https://github.com/onnx/onnx/actions/runs/7783887025/job/21232428145?pr=5906 ? Thanks!

> Quick question: do we want this in version 16? I thought it would be good to have, but if time does not allow I have no issue pushing it...

> Why can't the move happen up the stack earlier to clarify the ownership transfer? Makes sense too I think! Potential we can update the registration macros to use move...

I think it would be beneficial to include this which will allow us to do further validation on models. If anything comes up we can always revert the PR and...

We should.

One advantage with distributing the test data of course, is that runtimes do not need the Python tool chain to run tests (protobuf python, numpy etc.)

Maybe expect that given an input x, DFT(DFT(x, onesided=True), reversed=True, onesided=True)) == x? I think similar to https://numpy.org/doc/stable/reference/generated/numpy.fft.irfft.html