Open shaoyuyoung opened 2 months ago
Could you provide the simplified model for debugging?
sorry for the late response. below is the simplified model model-sim.zip In fact, the simplified model seems correct but the original model can't pass the compilation. BTW, the simplified tool I used is here: https://github.com/daquexian/onnx-simplifier
@xhmelon
Hello, sorry to bother you again but I still feel confused about this bug. @xhmelon
Is there any new progress on this issue?
I know maybe you have no time to investigate this because of your busy schedule.
I am trying to compile an ONNX (graph below) model using TVM.![5618520ff3e8817d39d4422c547eaf9](https://github.com/apache/tvm/assets/100203773/0a00c2d4-70e1-4e66-b7ae-96737e30d5b3)
Of course, this is a complicated graph, but we can simplify it as below.![image](https://github.com/apache/tvm/assets/100203773/10fd68b9-168f-4db7-b593-a88bd410e612)
These two graphs are equal. When I try to compile them using TVM. The original ONNX model fails but the simplified ONNX model passes. It is very strange!
This seems to involve the
Pad
operator shape-checking problem.In theory, I think TVM should have strong compatibility with the native ONNX model. However, the truth is not satisfactory.
It seems that only simplified, simple models are acceptable to TVM
Expected behavior
ONNX compilation passes
Actual behavior
Environment
Operating System: Ubuntu 18 TVM:0.15 Torch: 2.1.1 ONNX: 1.15.0
Steps to reproduce
ONNX file is here: onnx.zip
Here is the script
Triage