-
Notifications
You must be signed in to change notification settings - Fork 370
Closed
Labels
No Activitycomponent: convertersIssues re: Specific op convertersIssues re: Specific op converterscomponent: partitioningfeature requestNew feature or requestNew feature or request
Description
Unsupported operator: aten::atan2(Tensor self, Tensor other) -> (Tensor)
-
Function Schema:
aten::atan2(Tensor self, Tensor other) -> (Tensor) -
Original PyTorch API:
torch.atan2(input, other, *, out=None) → [Tensor] -
Relevant TensorRT Documentation:
None
Alternatives
Additional context
When using Fourier transformation in audio segmentation models, torch.atan2
will be used.
Because variable audio input lengths requires dynamic shape, fallback to pytorch would cause runtime error when creating example inputs by partitioning::generateRandomInputs
.
It reports
RuntimeError: Trying to create tensor with negative dimension -1: [1, 1, -1]
Therefore, we need convertor for torch.atan2
to successfully convert such models.
Metadata
Metadata
Labels
No Activitycomponent: convertersIssues re: Specific op convertersIssues re: Specific op converterscomponent: partitioningfeature requestNew feature or requestNew feature or request