export onnx error ( We don't have an op for aten::mul but it isn't a special case )

OpenDriveLab / TopoNet

Topology Reasoning for Scene Perception in Autonomous Driving

Apache License 2.0

280 stars 11 forks source link

export onnx error ( We don't have an op for aten::mul but it isn't a special case ) #17

Closed reskipper closed 1 month ago

reskipper commented 6 months ago

when i export toponet to onnx ，report error as below : RuntimeError: 0INTERNAL ASSERT FAILED at "../torch/csrc/jit/ir/alias_analysis.cpp":532, please report a bug to PyTorch. We don't have an op for aten::mul but it isn't a special case. Argument types: Tensor, bool,

Any suggestions on how I can solve this problem? Thanks a lot.

sephyli commented 6 months ago

Thanks for your question. You may can upgrade your PyTorch version to fix this. The repository's dependency is flexible.

reskipper commented 6 months ago

I tried torch 1.13.0+cu116 and mmcv-full 1.7.0, but the error persists. If I upgrade the torch version further, the mmcv-full version compatible with it may not be suitable for mmdet. If there are specific recommended versions, please let me know. Thanks!!

sephyli commented 6 months ago

I have a combination with torch 2.0 and CUDA 11.7/11.8, feel free to try. I think torch 1.13 can also compatible with mmcv-full 1.6.0. You can directly try on 1.13.

torch==2.0
torchvision==0.15.0
mmcls==0.25.0
mmcv-full==1.6.0
mmdet==2.26.0
mmdet3d==1.0.0rc6
mmsegmentation==0.29.1

reskipper commented 6 months ago

Thank you for your suggestion. I've tried it, but it still doesn't work. In fact, I encountered this error while exporting onnx for several repos. I will come back to provide further details if I find a solution.