Support for PyTorch torch.distributions.Normal

[itssimon]

Feature Request

Would be great if ONNX would implement support for the broadcast_tensors operator in PyTorch.

Currently torch.onnx.export() fails if the model uses torch.distributions.Normal:

RuntimeError: Exporting the operator broadcast_tensors to ONNX opset version 11 is not supported. Please open a bug to request ONNX export support for the missing operator.

Also see pytorch/pytorch#30517.

[mxkrn]

Same deal in opset_version=12.

Exporting the operator broadcast_tensors to ONNX opset version 12 is not supported. Please open a bug to request ONNX export support for the missing operator.

The new PyTorch distributions module makes extensive use of this method so it would be great to see support for this.

[Lenbus]

Have the same deal with PyTorch1.7 on every opset (9 to 12).

[rickstaa]

This feature and/or pytorch/pytorch#29843 are the only ones stopping us from moving over to Pytorch from Tensorflow. In Pytorch, I could not find an (easy) way to deploy any of our Gaussian and Baysian algorithms wherein Tensorflow is extremely easy.

[stale]

Is this still relevant? If so, what is blocking it? Is there anything you can do to help move it forward?

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.

[rickstaa]

I think this is still relevant.

[rickstaa]

Why didn't the bot re-open this 🤔?

[andife]

What is the current situation? I'm also interested in "torch.distributions.normal.Normal"

[rickstaa]

Maybe somebody of the @pytorch team can reopen this issue 🤔?

[rickstaa]

Thanks! 🙏

[justinchuby]

Replaced and tracked by pytorch/pytorch#111034