torch.istft should keep accepting real inputs in the future and not display a warning, because of ONNX #92204
Assignees
Labels
module: fft
module: onnx
Related to torch.onnx
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Comments
lezcano
added
module: onnx
|
We will be trying to gain support for new onnx ops that deal with complex numbers instead. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
馃殌 The feature, motivation and pitch
Right now when using torch.istft with float inputs (last dimension being [2]), the following warning is displayed:
UserWarning: istft will require a complex-valued input tensor in a future PyTorch release. Matching the output from stft with return_complex=Truehttps://github.com/pytorch/pytorch/blob/v1.13.1/aten/src/ATen/native/SpectralOps.cpp#L977-L981
However ONNX doesn't support (or support very badly) complex values. The ONNX STFT operator only output float values, with the last dimension being [2], and torch.stft with return_complex=True isn't supported when exporting to ONNX:
#92087
Therefore the future ONNX ISTFT operator is expected to take a float input with last dimension [2]. So torch.istft should remain compatible with that format in the future.
Alternatives
No response
Additional context
No response
cc @mruberry @peterbell10
The text was updated successfully, but these errors were encountered: