ONNX dynamic sized model export with torch.onnx.dynamo_export fails when torch.nn.functional.interpolate is used #124884
Labels
module: onnx
Related to torch.onnx
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
馃悰 Describe the bug
I'm trying to export a model from pytorch to onnx that needs to support dynamic input/output sizing at runtime. The model uses an interpolate step to resize features at a few points in the model. However, when
dynamic_shapes
option is set toTrue
, this results in a very long error that seems to indicate the Onnx export fails at the interpolate step. Here's a short example that declares an extremely simple model, verifies it will run a forward pass, but then fails on export.The error printout is quite long, but the most informative part is the following (Full error printout at end):
I've tried various
scale_factor
sizes, as well as different interpolation modes. However, they all seem to cause a similar export error. However, when I disabledynamic_shapes
in the export options, they work.Is there hope to get the interpolate operation supported for dynamic export?
Full error printout:
Versions
The text was updated successfully, but these errors were encountered: